An Exit Nearly a Decade in the Making

Achiam told staff on July 7 that he was stepping down, effective later this month. He cited no single reason, saying it had been building for a while. His note named no next step, and OpenAI hasn’t commented — a silence that stands out given how visible his role was.

Why does this departure matter? Before he held any title, Achiam had already built Spinning Up in Deep RL and helped write early safe-RL research, including work on constrained policy optimization. That’s why people listened to him on safety 

The Man Behind the Title

Achiam joined OpenAI back when it was still a nonprofit lab. Chatbots couldn’t hold a real conversation yet. Safety work defined most of his career there. He later led the mission alignment team, a group OpenAI dissolved earlier this year. He then took on a newly created title: chief futurist. The role sat at the intersection of safety, policy, and mission — studying AI’s risks as capabilities grew.

He was never a public face like Sam Altman. But insiders associated him with the harder, slower questions about AI’s trajectory — the kind that don’t make headlines but shape decisions behind the scenes.

Why This Exit Is Different

Most leadership departures blend into the background. This one drew attention for a reason. The chief futurist title never carried formal veto power over products or funding. Achiam’s influence came from being listened to, not from a line on an org chart. That kind of role is easy to quietly retire, and OpenAI hasn’t said whether it will.

Achiam himself didn’t frame the exit as a break from OpenAI’s mission. In his note to staff, he wrote that the world is now “in on the secret.” That, he suggested, makes it possible to keep pushing for safe AI from outside a frontier lab. In other words, he’s describing his time at OpenAI as finished, not abandoned.

A Timeline of Departures

Achiam isn’t the first senior safety-focused leader to leave OpenAI in recent years:

• 2024: Jan Leike departs for Anthropic. Miles Brundage and Steven Adler also leave.
• Late 2025: Andrea Vallone departs.
• February 2026: OpenAI dissolves its mission alignment team; Achiam moves into the newly created chief futurist role.
• July 7, 2026: Achiam tells staff he’s leaving, citing no single triggering event.
• The same week: Dean Ball, a former White House AI adviser, joins OpenAI in a related strategic futures capacity — a sign the company is still filling long-term thinking roles, even as Achiam’s own position sits open.

The Bigger Signal

This fits a broader industry pattern. Research labs that scale into consumer platforms tend to lose their safety-focused leaders first, often quietly. From Leike’s exit in 2024 to Achiam’s now, two years of departures have passed. They’ve tracked alongside OpenAI’s harder push into products and revenue.

Achiam’s exit closes a chapter that began at a small nonprofit lab. The mission statement he helped protect isn’t disappearing from the website — but the question is whether it still shapes decisions the way it used to. What’s worth watching now is simpler: who fills that empty chair, if anyone. A product-focused pick, or no replacement at all, would show OpenAI’s real priorities.

FAQs

What is Joshua Achiam known for?

Achiam joined OpenAI as an intern in 2017 and spent nearly nine years working on AI safety. He’s best known for creating Spinning Up in Deep RL, still used by RL researchers today. He also did early work on safe RL methods like constrained policy optimization.

Why did OpenAI create a “chief futurist” role in the first place?

OpenAI created the title for Achiam in February 2026, after dissolving the mission alignment team he led. The idea was to keep someone focused on AI’s long-term trajectory as the company shifted further toward products.

Should I be worried about AI safety at OpenAI?

One departure doesn’t answer that on its own. OpenAI says safety and policy work continues under other staff. Achiam himself framed his exit as continuing the same goals from outside the company, not breaking from them.

Is Joshua Achiam joining a competitor?

His note didn’t name a next step, and nothing has been confirmed publicly. OpenAI hasn’t issued an official comment either.

Will OpenAI replace the chief futurist role?

OpenAI hasn’t said. Because the title was created specifically for Achiam, there’s no guarantee it continues without him.

The post OpenAI’s Chief Futurist Is Leaving. What Changed After Nine Years? appeared first on Memeburn.

What the Certification Image Actually Shows

Safety Korea published a certification record for a tablet carrying model numbers SM-X840 and SM-X846N, according to 91 Mobile. The SM-X846N variant is the Korean model with built-in 5G connectivity.

The image itself doesn’t reveal groundbreaking details. It’s clearly a Samsung tablet, and it keeps the same flat-edged, minimal-bezel design language we’ve seen across the Tab S series for the past three years. No redesign. No surprise form factor shift. If you placed them side by side with a Tab S10+ or Tab S11, you’d have a hard time telling them apart from the front alone.

But here’s what matters: certifications like this typically surface about two months before a device hits shelves. With Samsung’s Galaxy Unpacked 2026 event happening on July 22 and a separate September tablet launch expected, the Tab S12+ is clearly on schedule.

Why Samsung Is Dropping the Base Model Again

The bigger story isn’t the image — it’s what’s not in the lineup. Samsung is skipping the base 11-inch Galaxy Tab S12 entirely, offering only the 12.4-inch Tab S12+ and the 14.6-inch Tab S12 Ultra.

This isn’t random. Samsung has settled into an alternating pattern: in 2024, it released the Tab S10+ and Tab S10 Ultra. In 2025, it brought back the 11-inch Tab S11 and Ultra, but dropped the Plus. Now the Plus returns and the base disappears again.

For buyers, this matters because it raises the floor price of Samsung’s flagship tablet lineup by roughly $300. If you were previously entering the Tab S ecosystem at $799 with the base model, that door is closed this year. The Tab S12+ is expected to start between $1,100 and $1,200 according to ETNews reporting — pricing that puts it toe-to-toe with the iPad Pro, not the iPad Air.

In our view, Samsung is making a deliberate bet. It’s positioning its flagship tablets as laptop alternatives, not media consumption devices. Whether buyers agree with that repositioning is the question that matters most.

What We Know About the Hardware

No specs surfaced from the Safety Korea filing itself, but months of leaks have painted a fairly detailed picture. Both the Tab S12+ and Tab S12 Ultra will run on the MediaTek Dimensity 9500, a 3nm chipset confirmed through an APK teardown of Samsung’s AI Core app.

That processor represents a genuine generational leap. Compared to the Tab S11’s Dimensity 9400+, the 9500 delivers up to 32% faster single-core CPU performance, 119% faster ray tracing, and roughly double the AI compute power — all while drawing 55% less power at peak CPU load.

Here’s a quick look at what’s expected:

One detail worth flagging: the Tab S12 Ultra’s battery stays identical to last year’s model at 11,600mAh. Samsung appears to be counting on the Dimensity 9500’s power efficiency gains to stretch battery life without increasing cell size. That’s a reasonable engineering tradeoff, but it’s also the kind of decision that makes a $1,200 “upgrade” harder to justify for current Tab S11 Ultra owners.

When You Can Actually Buy It

Samsung’s September release window for flagship tablets has become predictable by now. The Tab S11 series launched on September 4, 2025. The Tab S10 dropped in September 2024. IFA 2026 runs September 4 through 8 in Berlin, making it a natural stage for the announcement.

If Samsung follows its typical cadence, the Tab S12+ and Tab S12 Ultra should be available for purchase within two weeks of the announcement. That puts the earliest possible availability around mid-to-late September — perfectly timed for the holiday shopping season and any Black Friday deals that follow.

Samsung’s recent AI-driven profit surge shows the company is investing heavily in chips and artificial intelligence across its product lines. The Tab S12 series will likely lean hard on Galaxy AI features — including on-device image generation, smarter call screening, and expanded DeX productivity tools — to justify that premium pricing.

The Bottom Line

This certification image confirms what we’ve suspected for months: the Galaxy Tab S12+ exists, it’s real, and it’s on track for a September launch. Samsung isn’t reinventing the wheel with its design, and the spec upgrades — while meaningful on the chipset side — don’t represent the kind of generational leap that makes existing Tab S owners rush to upgrade.

But if you’re coming from an older tablet or shopping for your first premium Android slate, the Tab S12+ looks like a solid entry point. Just be prepared for the price tag.

FAQs

How does the Galaxy Tab S12+ compare to the iPad Pro 2026? 

Both target the same $1,100+ price bracket, but they run different ecosystems. Samsung’s OLED display technology and S Pen are strong selling points, while the iPad Pro offers Apple’s M-series chip and a more mature tablet app library. Your choice depends on whether you prefer Android’s flexibility or iPadOS’s polish.

Will Samsung Galaxy Tab S12+ work with existing S Pen accessories? 

Samsung has historically maintained backward compatibility with S Pen accessories across Tab S generations. The Tab S12+ is expected to ship with an S Pen in the box, and older Samsung tablet accessories like keyboard covers should remain compatible, though Samsung hasn’t confirmed this officially.

What Android version will the Galaxy Tab S12+ ship with? 

The Tab S12+ is expected to launch with Android 17 and Samsung’s One UI 9.0. Under Samsung’s current update policy, it should receive seven years of OS upgrades — meaning software updates through Android 24 and security patches through 2033.

Is the MediaTek Dimensity 9500 as good as Qualcomm Snapdragon chips? 

For tablets, MediaTek’s premium chips actually have an edge. The Dimensity 9500’s thermal management suits thin tablet form factors better than Qualcomm’s hotter-running designs, and Samsung’s three-generation commitment to MediaTek in its tablets suggests the partnership is working well.

What happened to Samsung Galaxy Unpacked 2026? 

Samsung confirmed its Galaxy Unpacked event for July 22 in London, where it’ll unveil new foldables, smartwatches, and Galaxy Glasses. The Tab S12 series won’t appear at Unpacked — it’s being held for a separate September announcement, likely timed around IFA Berlin.

The post Samsung Galaxy Tab S12+ Just Showed Up in Its First Live Image appeared first on Memeburn.

This isn’t just a smoother chatbot. For South African readers, the bigger question is whether voice AI can become useful across support lines, classrooms, banking apps and multilingual daily life without becoming another expensive feature locked behind premium plans.

What OpenAI actually released

On 8 July 2026, OpenAI introduced GPT-Live, a family of voice models that now powers ChatGPT Voice. The company says the models make talking with AI feel closer to a real-time conversation instead of the old “you talk, it waits, then it replies” rhythm. 

That sounds small until you use voice tools in the real world: background traffic, half-finished thoughts, and those awkward silences where software thinks you’re done.

OpenAI says more than 150 million people use ChatGPT features like Voice and Dictation each week. We think the real story here is not the voice itself. It’s that OpenAI wants ChatGPT to become something you talk to while it searches, reasons, checks context, and brings visual answers back to your screen.

Why “full-duplex” matters

The key technical change is full-duplex audio. Plainly, that means GPT-Live can listen and speak at the same time, like a person on a phone call. Older systems treated speech as neat turns. You spoke, they waited, then they answered. If you paused for one second, they often jumped in. 

OpenAI says GPT-Live continuously processes speech while generating output, making decisions many times per second about whether to talk, listen, pause, interrupt, or use a tool.

That should make everyday conversations feel less stiff. You can ask ChatGPT to slow down, interrupt with a follow-up, or pause while you think. It can respond with listening cues such as “mhmm” or “got it,” although we’ll need to see how natural that feels across accents and languages. 

The upside is fewer bot-like call-centre moments. The risk is that a warmer voice may make users over-trust the tool.

The new ChatGPT Voice, at a glance

Why this could matter in South Africa

For South Africa, the obvious use case is customer service. Banks, insurers, telecoms and retailers already spend heavily on support channels. A voice agent that understands interruptions, background noise and messy requests could cut frustration for users who don’t want to type a long complaint. 

But trust will matter. A banking voice bot that sounds friendly still needs clear escalation to a human, strong privacy controls, and plain warnings when it doesn’t know something.

Education is another area to watch. A more natural voice tutor could help learners practise languages, revise exam topics, or ask questions without perfect spelling. South Africa’s AI policy framework already highlights the digital divide and points to education, finance, health, agriculture and public governance as important AI sectors. Voice could lower the first barrier: typing.

But affordability remains the awkward part. OpenAI says GPT-Live-1 will become the default for Go, Plus and Pro users, while GPT-Live-1 mini will power ChatGPT Voice for Free users. That split matters in markets like South Africa, where exchange rates and subscription fatigue shape who gets the best tools first. If the free version feels weaker, the gap between casual users and paying users could widen.

Safety is now part of the product

OpenAI also says GPT-Live includes safeguards that can act during a live voice conversation. The system can steer away from unsafe output, add safety resources, or end a conversation in higher-risk cases. It also includes protections for teen users and predefined voices designed to avoid real-person impersonation.

This matters because voice feels intimate. Text gives you distance. A calm, responsive voice can feel more persuasive and more present. That’s useful when you’re learning or navigating a task. It’s risky when the topic involves mental health, money, identity, or personal crisis. What we’re watching now is how OpenAI balances warmth with boundaries.

The bigger AI strategy

This release also fits OpenAI’s bigger push to turn ChatGPT from a chatbot into a broader AI workspace. Our recent Memeburn breakdown of OpenAI’s ChatGPT overhaul tracked the company’s move toward agents, coding tools and a more app-like experience. GPT-Live gives that strategy a natural interface: you talk to the system while it works.

The interesting part isn’t just that ChatGPT can now sound more natural. It’s what that says about the next phase of consumer AI. 

The winning AI apps may not be the ones with the longest feature lists. They may be the ones you can use while walking through Cape Town, riding the Gautrain, cooking dinner, or helping a child with homework. So, when AI starts sounding less like software and more like a person, will you use it more — or trust it less?

FAQs

What is OpenAI GPT-Live?

GPT-Live is OpenAI’s new voice model family for ChatGPT Voice. It can listen and speak at the same time, which should make chats feel more natural.

Who gets GPT-Live?

OpenAI says GPT-Live-1 is rolling out globally for Go, Plus and Pro users. Free users get GPT-Live-1 mini as the default ChatGPT Voice model.

Why does GPT-Live matter for South Africa?

It could make AI easier to use in customer support, education and multilingual access. The real test is affordability, privacy and how well it handles local accents and languages.

The post OpenAI GPT-Live makes ChatGPT voice chats feel natural in 2026 appeared first on Memeburn.

What’s New in Instagram AI

(30 new effects in Stories powered by Muse Image. Source: Instagram)

Meta just handed strangers a shortcut into your camera roll, without asking. Its new Muse Image tool, rolling out in beta, lets anyone reference a public Instagram profile in an AI prompt and pull from that account’s real photos. It’s the latest addition to Meta’s in-house AI stack, alongside Imagine, already in the Meta AI app.

The feature is on by default for public accounts, so a five-year-old selfie you forgot existed can now feed someone else’s AI-generated creation. It’s live in the US first, with more countries expected to follow.

This isn’t a feature you opt into. It’s one you have to notice and shut off yourself, in a menu most people have never opened.

How Muse Image Actually Works

Meta describes the tool as turning simple prompts into finished visuals. People can edit existing photos or generate new ones from scratch. Someone can @-mention a public profile in a prompt. The model then pulls from that account’s photos to build the image.

This differs from AI training, where data shapes a model’s general behavior. Muse Image is closer to remixing: it grabs your specific photos on demand, one image at a time.

The Opt Out Nobody Asked For, But Everyone Needs

Here’s the fix if going private isn’t an option:

• Open Instagram and tap your profile
• Tap the three-line menu, top right
• Scroll to Sharing and reuse
• Find “Allow people to use your content on Instagram and with AI features on Meta”
• Turn off the toggles for both Posts and Reels

Two catches. The on/off states look nearly identical, so double-check it landed right. And this only stops future generations, not past ones.

How The Internet Reacted

Most people reacting to this made the same basic point: Instagram was always public, so anyone could already screenshot or save those photos. In their view, Muse Image isn’t really new, it’s just an official version of something that was already possible.

That reaction works in Meta’s favor. When people shrug and say “that’s fair enough” instead of pushing back, there’s no pressure on Meta to change course. And when a feature doesn’t get real resistance, it doesn’t just avoid criticism, it starts to feel normal.

Why This Matters Beyond Instagram

Meta has used public Facebook and Instagram content to train its AI models since 2024. That policy is contestable mostly in the EU or UK. Muse Image differs: instead of a static model, it pulls your specific photos into a live, one-off creation in real time.

Meanwhile, there’s a business angle. Meta reportedly built the tool in-house after previously licensing Midjourney. It’s already positioned to plug into ad products like Advantage+.

For influencers and brand accounts, that’s the real story. Visibility is the basis of the creator economy. Now it doubles as free material for Meta’s own commercial AI, with no fee or credit. Going private isn’t realistic when your account is your business.

The Takeaway

Meta just turned a settings menu into the only defense against someone else’s AI prompt. If your account is public, your content already has a role in this system, whether you meant it to or not. For creators and brands, checking that toggle is basic upkeep now.

FAQs

Can Meta use my photos for AI?

Yes, depending on your settings. If your account is public and you haven’t opted out, any post or reel can serve as visual input for someone else’s Muse Image prompt. There’s no separate consent step involved.

Does this affect private accounts?

No, only public profiles are included by default. Private accounts sit outside Muse Image entirely, so switching to private removes your content, though that’s not realistic for creators or brands who need public visibility.

Do I get notified if my photo is used?

Reportedly, no. Meta’s help documentation is said to state users won’t get any alert when their content is used. That means you’re left assuming it’s possible, rather than confirming it happened.

How do I stop it?

Go to Settings > Sharing and reuse. Turn off the toggles for both Posts and Reels, or switch to a private account. Note that this only affects future use, not anything already generated.

Is this rolled out everywhere?

Not yet. Muse Image is currently in beta and live in the US only, with Meta expected to expand it to more countries over time.

The post Meta’s Muse Image Is Turning Old Instagram Photos Into New AI Experiences appeared first on Memeburn.

What Muse Image Actually Does

Muse Image is Meta’s answer to OpenAI’s GPT Image and Google’s image generation tools. Built by Meta Superintelligence Labs under Chief AI Officer Alexandr Wang, the model creates images from text prompts, edits specific regions of existing photos, and powers over 30 AI visual effects across Instagram and WhatsApp.

Introducing Muse Image and Muse Video, the first media generation models developed by Meta Superintelligence Labs.

Muse Image is our most advanced image generation model yet. It follows instructions faithfully, edits with precision, composes from multiple references, and draws… pic.twitter.com/byNpQZO1RW

— AI at Meta (@AIatMeta) July 7, 2026

The part that sparked immediate backlash, though, is the @-mention feature. Type someone’s public Instagram handle into a Meta AI prompt, and the system pulls their photos as visual references to generate new images featuring their likeness. No permission request. No notification. Meta’s own help page says it plainly: you won’t be told when someone creates content using your photos.

That’s a significant shift. We’ve seen AI tools scrape the open web for training data before. But this is different — it’s a feature explicitly designed to let one user manipulate another user’s identity, and Meta built it as the default experience.

The Opt-Out Problem

Meta says users “remain in control” through privacy settings. Technically true. Practically misleading. The opt-out path requires four steps: open Instagram, tap your profile, navigate to Sharing and reuse, and toggle off both Posts and Reels under the AI features section.

There are two catches worth highlighting. First, opting out only stops future AI generations. Any images someone has already created using your photos stay in the wild. Second, some users couldn’t even see the updated settings language on launch day, meaning the controls weren’t visible yet, while the feature was already live.

Meta did confirm that private accounts and users under 18 are automatically excluded. But for the roughly 2 billion public Instagram accounts worldwide, the default is: you’re in.

We’ve seen this pattern before. When Meta began using EU users’ public posts to train its Llama models in May 2025, the company relied on the same opt-out approach — and privacy group NOYB immediately threatened legal action, arguing that GDPR requires opt-in consent. A German court ultimately sided with Meta on the AI training question. But generating recognizable images of specific people from their social media photos is a fundamentally different use case than training a language model on aggregated text.

Why This Isn’t Just a Privacy Settings Story

The deeper issue here is what “public” means in the context of AI. When you set your Instagram account to public in 2015, you were consenting to people viewing your photos. You were not consenting to a system that lets a stranger generate synthetic images of your face in any context they choose.

That distinction matters. For creators, influencers, and anyone whose identity carries commercial value, AI-generated likenesses raise real risks: impersonation, fake endorsements, reputational damage, and brand dilution. Meta’s own history — from the Cambridge Analytica scandal that cost it a $5 billion FTC fine to the shutdown of its facial recognition system in 2021 — makes the default-on approach feel like a deliberate gamble rather than an oversight.

It also fits a broader pattern we’ve been tracking across the industry. Google Photos recently enabled AI image generation from users’ private libraries by default. Meta is now doing the same with public profiles at an even larger scale. The platforms that used to be neutral publishers of user content are quietly converting that content into fuel for generative AI — and they’re betting most people won’t notice until it’s too late.

This is also happening while Meta pours record capital into AI infrastructure, raising its annual CapEx forecast to between $125 billion and $145 billion. Muse Image isn’t just a creative tool — it’s wired directly into Meta’s advertising pipeline, where businesses can generate ad visuals using the same model. META stock climbed 2.17% to roughly $613 on launch day. The market clearly sees the commercial logic. Whether users see the same value is a different question.

Content Seal and the EU Problem

Alongside Muse Image, Meta introduced Content Seal — an invisible, compression-resistant watermark designed to help identify AI-generated images. There’s even a public detection tool in preview. On paper, that’s a step toward responsible AI deployment, and it mirrors the kind of AI safety infrastructure that Meta has been building through recent acqui-hires.

But “invisible” is the operative word. Starting August 2, 2026, the EU AI Act’s Article 50 requires anyone publishing AI-generated content in EU markets to visibly label it. An invisible watermark that’s machine-readable but not viewer-readable lands on the wrong side of that line. Whether Meta will add visible labeling for EU users hasn’t been announced. That’s not a small gap — it’s a compliance question that could determine whether Muse Image ships as-is in Europe next month.

What You Should Do Right Now

If your Instagram account is public, open your settings and turn off content sharing with AI features immediately. It takes 30 seconds. That won’t undo anything already generated, but it stops future use.

Beyond that, this is a moment to think about what “public” actually means on platforms where the rules keep changing. Tools like self-hosted AI workspaces exist specifically because the tradeoff between convenience and control keeps shifting in one direction. Meta’s move this week made that tradeoff harder to ignore.

FAQs

How does AI image generation differ from AI model training? 

AI model training uses large datasets to teach a system patterns and language. AI image generation takes a specific prompt and creates new visual content on demand. Training is a background process; generation produces an output someone can share or publish immediately. The distinction matters because consent expectations differ for each.

What is the EU AI Act and how does it regulate AI content? 

The EU AI Act is a comprehensive regulatory framework that took effect in stages starting 2024. From August 2026, it requires visible labeling on AI-generated content published in EU markets. Companies that fail to comply with EU AI Act transparency rules face significant fines. It covers chatbots, image generators, deepfakes, and more.

Can AI-generated images be used for identity theft or fraud? 

Yes. Synthetic images of real people can be used for fake endorsements, impersonation scams, phishing attacks, and social engineering. When AI tools make it easy to create realistic images of someone without their consent, the risk of identity-based cyber threats rises sharply. 

How do AI watermarks like Content Seal actually work? 

AI watermarks embed invisible signals into generated images that survive cropping, compression, and screenshots. Detection tools read these signals to verify whether an image was AI-made. However, invisible watermarks don’t replace visible AI content labels required by some regulations, and they can be defeated by sufficiently motivated actors. 

What privacy tools exist to protect your data from Big Tech AI? 

Self-hosted AI platforms like Odysseus let you run AI models locally without sending data to cloud servers. You can also adjust privacy settings on platforms like Instagram, disable AI training toggles, and use privacy-focused AI alternatives that don’t collect or retain user data.

The post Meta Instagram AI Photos: Muse Image Uses Your Face by Default appeared first on Memeburn.

What the Leaks Actually Say

The details come from Digital Chat Station, a well-known Weibo tipster with a reliable track record on Xiaomi hardware. In his latest post, he described a new Redmi performance phone slated for January with a 7-inch 2K flat display, a Dimensity 9-series chipset, and a battery in the 10,000mAh class. He didn’t name the device, but every detail matches earlier Redmi Turbo 6 Max leaks from May.

Back then, the same tipster revealed the phone would use a 3nm processor and pack an ultrasonic in-display fingerprint scanner, a metal middle frame, and IP68/IP69 water resistance. Those aren’t mid-range features. Their flagship-grade durability specs are being dropped into a phone that’ll likely cost half as much as a Galaxy S-series.

If you’ve followed Redmi’s recent product rollout — the Note 17 in July, the K100 in August — this fits a clear pattern. Redmi is stacking its lineup with increasingly ambitious hardware, and the Turbo 6 Max sits at the very top of that ladder.

Why a 7-Inch 2K Screen Changes the Category

The Turbo 5 Max already pushed limits at 6.83 inches. Going to 7 inches with 2K resolution (roughly 340+ ppi) doesn’t just make the screen bigger — it repositions the device entirely.

At this size, you’re carrying a compact tablet that makes calls. That’s not accidental. Redmi isn’t chasing the compact flagship crowd. The Turbo line targets users who want maximum screen real estate and endurance without paying flagship prices. A 7-inch chassis tells you exactly who this phone is for: people who binge content, game hard, or treat their phone as a primary work device.

For context, the Xiaomi 17T Pro’s 6.83-inch panel was already among the largest in its class this year. The Turbo 6 Max leapfrogs it to become the biggest non-foldable Redmi phone we’ve seen leaked.

The 3nm Chip Is the Biggest Wildcard

We know it’s a Dimensity 9-series processor built on a 3nm node. We don’t know which one — and that matters more than it sounds.

MediaTek currently offers the Dimensity 9400, 9400+, and the newer 9500, all on TSMC’s 3nm process. The Dimensity 9500 — announced in late 2025 — packs an upgraded Arm Mali-G1 Ultra MC12 GPU, a massive 19MB+ cache, and is built for sustained workloads and AI-assisted frame synthesis. If the Turbo 6 Max lands the 9500, it would match the silicon inside phones costing twice as much.

If it gets a trimmed variant like the 9500s (which powered the Turbo 5 Max), it’s still a strong performer — just not absolute flagship tier. Either way, MediaTek’s 3nm trajectory means this phone will handle heavy gaming and multitasking comfortably. The real question is how aggressively Redmi prices it against Qualcomm-equipped competitors.

Battery Tech That’s Actually Different

A 10,000mAh battery is already massive. But a separate April leak from Digital Chat Station puts the figure at 12,000 mAh in active subsystem testing — using a single-cell silicon-carbon anode design.

Here’s why that’s significant. Most ultra-large phone batteries above 8,000mAh use dual-cell configurations: two smaller cells stacked to manage heat and thickness. A single-cell 12,000mAh design at this density is genuinely novel. It produces a flatter battery profile that’s easier to fit inside a slim body.

For comparison, the iQOO Z11i launched at $190 with a 7,500mAh battery as its headline feature. The Turbo 6 Max could nearly double that. If Redmi delivers, the conversation shifts from “lasting a full day” to “how many days between charges.”

Competition and Conclusion

The Turbo 6 Max is expected to compete head-on with the OnePlus Turbo 7, rumored for the same January 2027 window in China with a Snapdragon 8-series chip. That matchup is telling — OnePlus has traditionally leaned on Qualcomm silicon and OxygenOS, though the recent merger into OPPO’s ColorOS has shifted its software identity. Redmi, meanwhile, bets on MediaTek and HyperOS. The Turbo 6 Max vs. Turbo 7 fight will come down to battery and pricing — two areas where Redmi typically has the edge.

Globally, expect the Turbo 6 Max to rebrand as the Poco X9 Pro Max, following the same localization pattern Xiaomi uses to reach markets outside China.

Redmi isn’t just building another phone. It’s building a category. The Redmi Note 17 Pro Max covers 10,000mAh at mid-range pricing. The K100 Pro handles flagship performance at 9,000mAh. The Turbo 6 Max goes above all of them. That’s not random — it’s a roadmap statement. Every buyer has a clear path, and none of Redmi’s own devices cannibalize each other. It’s smart product management, and it’s the part of this leak that deserves the most attention.

FAQs

What chipset does the Xiaomi 17T Pro use? 

The Xiaomi 17T Pro runs the MediaTek Dimensity 9500, the same chip family rumored for the Turbo 6 Max. It’s built on TSMC’s 3nm process and features the Arm Mali-G1 Ultra MC12 GPU with 19MB+ cache for sustained performance. You can read our full breakdown of the Xiaomi 17T Pro specs for more details.

How does the Redmi Note 17 Pro Max compare on battery? 

The Redmi Note 17 Pro Max packs a 10,100mAh battery with a Dimensity 7500 chip, targeting the mid-range segment. It’s positioned below the Turbo 6 Max in Redmi’s lineup. Our Redmi Note 17 series breakdown covers all three variants and their specs.

What is silicon-carbon battery technology? 

Silicon-carbon anodes replace traditional graphite in lithium batteries, allowing cells to store more energy in the same physical space. This is the tech behind the Turbo 6 Max’s rumored 12,000mAh single-cell design. Several 2026 phones, including models from iQOO and OnePlus, already use earlier versions of this chemistry.

Will the Poco X9 Pro Max have the same specs as the Turbo 6 Max? 

Historically, Xiaomi rebrands its China-only Redmi phones under the Poco label for global markets. Core hardware typically carries over — display, chipset, battery — but camera configurations and software features sometimes differ. The Poco X9 Pro Max is expected to mirror the Turbo 6 Max’s key specs.

What’s the difference between IP68 and IP69 water resistance? 

IP68 means a device can withstand submersion in water beyond 1 meter for extended periods. IP69 adds resistance to high-pressure, high-temperature water jets — originally designed for industrial equipment. Dual IP68/IP69 certification on a smartphone at this price point is uncommon and signals serious durability focus.

The post Redmi Turbo 6 Max Leak With 7-Inch 2K Screen, 3nm Chip appeared first on Memeburn.

Franklyn Wang, CEO of the trading platform Liquid and its AI product Co-Invest, submitted the winning answer on July 6. He identified the document as a December 2024 rewrite of EIP-7503, a proposal known as Zero-Knowledge Wormholes, which had been submitted through the anonymous GitHub account 0xwormhole. Buterin responded on X with a simple acknowledgment of the result.

How the Experiment Started

Buterin launched the challenge on June 22 with a post on X. He stated that claims about AI text analysis making online anonymity untenable had prompted him to test the idea against his own record. He revealed that at some point during the past decade, he had written and published an Ethereum-related document under a different identity and challenged anyone with access to AI tools to find it.

He described the document as carrying “medium importance” to Ethereum and estimated it sat among roughly 200 to 2,000 publications of similar or greater significance within the ecosystem. He provided no hints about the title, publication venue, or the pseudonym he used.

The framing was intentional. Buterin has one of the most extensive publicly available writing records in crypto, spanning blog posts, Ethereum Improvement Proposals, research papers, forum comments, and social media. If AI could not identify his anonymous work given that volume of reference material, the experiment would suggest pseudonymous contribution remains durable. If it could, the implications would run in the opposite direction.

What AI Found and How

Wang’s approach relied on Co-Invest, a ChatGPT and Claude-compatible AI trading tool that Liquid launched in May 2026. Rather than building a custom pipeline, Wang applied Co-Invest’s analytical capabilities to a corpus of 27 Ethereum-related documents.

The system ranked Buterin as the most likely author of the anonymous EIP-7503 rewrite with approximately 20% confidence. That figure may sound low, but it was roughly 10 times higher than the probability assigned to any other candidate in the analysis. Across a pool of 27 documents and potentially dozens of candidate authors, that margin was sufficient to single out Buterin with a correct conclusion.

What made the identification notable was not the confidence level but the signal the AI latched onto. Wang stated that the tool did not detect Buterin through vocabulary, sentence structure, or any surface-level writing feature. Instead, it identified recurring patterns in how the document explained mathematical concepts, cryptographic reasoning, and algorithmic logic. These intellectual habits, as Buterin himself called them, functioned as a cognitive fingerprint that survived his deliberate attempts at disguise.

Buterin’s Failed Disguise Strategy

Buterin later revealed the method he had used to mask his authorship. He wrote the entire document in Chinese, used the Qwen 2.5 language model running locally to translate it into English, and then manually corrected errors in the translation to remove remaining traces of his usual prose style.

The strategy targeted surface-level stylometric markers like word choice, sentence length, and grammatical patterns. Against traditional text analysis, this approach might have been effective. Against an AI system capable of evaluating how ideas are structured and how technical arguments flow, it fell short.

Buterin acknowledged this gap in his response, noting that the AI had detected his “intellectual habits and style of math and algorithm explanation,” which bypassed his obfuscation strategy completely. The strategy only addressed prose, leaving the deeper layer of reasoning exposed.

This suggests a distinction that most existing discussions about AI and anonymity have overlooked. Surface-level text disguise may protect against older stylometric tools that match vocabulary and syntax. But AI systems trained on large corpora can now evaluate the structure of technical arguments at a deeper level, where the author’s way of thinking becomes the identifying marker rather than their way of writing. For technical contributors in particular, this is far harder to disguise because it requires not just changing words but changing how you reason through problems on the page.

The Research That Prompted the Experiment

Buterin’s challenge did not emerge in isolation. In February 2026, researchers from ETH Zurich and Anthropic published a paper titled “Large-scale online deanonymization with LLMs” that demonstrated how language models can automate the identification of pseudonymous users at scale. The study has not yet been peer-reviewed, but its findings were striking.

The research team built an automated pipeline with four stages. It extracted identity-relevant features from unstructured text, searched for candidate matches using semantic embeddings, reasoned over the top candidates, and assigned confidence scores. In tests linking Hacker News accounts to LinkedIn profiles, the system achieved up to 68% recall at 90% precision, far outperforming classical deanonymization methods that require structured data.

The researchers estimated the cost at roughly $1 to $4 per identification, making the technique accessible to any moderately resourced entity. Lead researcher Simon Lermen noted that the system was automating what skilled human investigators could already do manually, but at a fraction of the time and cost.

Why This Matters for Blockchain

Pseudonymous contribution is fundamental to how open-source blockchain communities operate. Developers, security researchers, and protocol contributors regularly use pseudonyms to reduce personal exposure, avoid conflicts of interest, or protect themselves from targeted attacks. The most famous example remains Bitcoin’s creator, Satoshi Nakamoto, whose identity has resisted identification for over 15 years.

Wang himself has prior experience in this space. Lighter CEO Vladimir Novakovski confirmed that he and Wang ran a 2023 project using GPT-4 to analyze writing style across cryptography research papers, attempting to identify Satoshi. That effort did not produce a high-confidence result, which may suggest that AI deanonymization works better against living authors who produce ongoing public writing than against historical figures with limited and static text samples.

The practical implication is that contributors who continue to write technical documents, even under pseudonyms, create an expanding corpus that AI systems can compare against. Each new publication adds data points that strengthen potential identification, particularly when the contributor has a distinctive way of approaching technical problems. Buterin’s case illustrates this directly. He has published millions of words over more than a decade, giving AI systems extensive training material to work with.

The risk is not evenly distributed. High-profile contributors like Buterin have the largest public writing footprints and the most distinctive reasoning styles. But the ETH Zurich study showed the technique works against ordinary users as well, linking pseudonymous forum accounts to real identities using only post histories.

What Remains Unclear

Several questions remain open after the experiment. Buterin’s test involved a single document from a single well-known author with an unusually large body of public writing. Whether the same approach would work against less prolific or less distinctive writers has not been tested in this context.

The 20% confidence score also warrants scrutiny. Wang’s identification was correct, but the system was far from certain. In a scenario without Buterin’s confirmation, a 20% probability would typically be treated as a lead worth investigating rather than a definitive conclusion. The strength of the result came from the relative margin over other candidates, not from absolute certainty.

Privacy on the Lean Ethereum roadmap has been elevated to a first-class protocol goal, with Buterin’s own plans covering transaction privacy, mempool confidentiality, and metadata reduction. But those upgrades address on-chain privacy. Authorship privacy occupies a separate domain that protocol changes cannot directly protect.

The next meaningful test would involve a less prominent author, a smaller reference corpus, and no public confirmation. Until those conditions are tested, the experiment demonstrates that AI can deanonymize a highly distinctive technical writer. Whether it can do so reliably across the broader population of pseudonymous blockchain contributors remains an open question.

FAQs

What document did Vitalik Buterin write anonymously?

Buterin authored an anonymous rewrite of EIP-7503, titled Zero-Knowledge Wormholes, submitted in December 2024 through the GitHub account 0xwormhole. He confirmed this after Franklyn Wang’s AI-assisted analysis correctly identified the document.

How did AI identify Vitalik Buterin as the anonymous author?

Co-Invest’s AI analyzed 27 documents and identified Buterin through his reasoning patterns, mathematical explanations, and cryptographic argumentation style. Surface-level prose features like vocabulary and grammar were not the primary signals.

What is AI deanonymization and why does it matter for crypto?

AI deanonymization uses large language models to match anonymous writing to known authors by analyzing text patterns at scale. It matters for crypto because many blockchain developers, researchers, and protocol contributors rely on pseudonyms for security and privacy.

Can AI identify Satoshi Nakamoto using the same technique?

Franklyn Wang and Vladimir Novakovski attempted a similar approach in 2023 using GPT-4 to analyze cryptography research papers, but the effort did not produce a high-confidence match. Satoshi’s limited and static text corpus makes identification harder than active writers like Buterin.

The post Vitalik Buterin Confirms AI Cracked His Anonymous Ethereum Work appeared first on Memeburn.

What the Samsung Galaxy A18 Renders Show

If you put the Galaxy A18 next to the outgoing Galaxy A17, you’d struggle to tell them apart. The phone keeps its 6.7-inch Infinity-U display — that’s the U-shaped notch at the top for the selfie camera — along with thick bezels that are standard at this price. Samsung also retains its “Key Island” frame design, where the power and volume buttons sit on a slightly raised strip along the side.

On the back, the triple-camera setup returns in a vertical pill-shaped housing. The one visible tweak: the camera module now sits a little flatter against the rear panel, giving the phone a cleaner look compared to the A17’s more protruding bump.

Dimensions tell the same story. The Galaxy A18 measures 164.4 × 77.8 × 7.84mm — just 0.1mm narrower and 0.34mm thicker than its predecessor. You won’t feel that difference in your hand. Whether the extra thickness means a bigger battery remains unclear, though current rumors still point to a 5,000mAh cell.

Samsung could’ve upgraded the display to a punch-hole cutout — it already did that for the Galaxy A27 this year. The fact that it didn’t signal a deliberate cost decision, not a design oversight. At this price point, Samsung’s priority is keeping production costs low enough to protect margins while memory chip prices squeeze budgets across the industry. It’s a boring choice, but it’s a rational one.

Why Samsung Is Dropping Exynos for Snapdragon

The real story is the chipset. According to a ZDNet Korea report from July 1, the Galaxy A18 5G will ditch Samsung’s own Exynos processor in favor of a Qualcomm Snapdragon chip — likely the Snapdragon 6s Gen 3. The 4G variant sticks with MediaTek, probably a Helio G99.

That’s a notable strategic shift. Samsung has traditionally used Exynos silicon across much of its A-series lineup, especially in markets outside the U.S. Moving to Snapdragon for its entry-level 5G phone suggests Samsung is prioritizing consistent modem performance and power efficiency over cost savings from using its own chips.

Both processors are entry-level parts. But Snapdragon chips generally offer better 5G modem integration and more optimized power management — two things that matter a lot when your target buyer cares about battery life and connectivity above all else. The Galaxy A18 4G has already appeared in the GSMA database as SM-A185F, confirming that Samsung is actively preparing both models for launch.

This isn’t just a spec swap. If Samsung rolls Qualcomm silicon into its budget 5G phones more broadly, it changes competitive dynamics in the segment. Exynos chips have been criticized for inconsistent performance and thermals, particularly in emerging markets where network infrastructure varies. Moving to Snapdragon is a quiet concession — Samsung is essentially saying its own chips aren’t the right fit here.

Samsung’s Budget Phones Outsell Its Flagships

Here’s why a phone you’ve never heard of matters more than you’d think. According to Counterpoint Research, Samsung placed five phones on the global top 10 best-sellers list for Q1 2026 — and every single one was a budget A-series model. The Galaxy A07 4G was the best-selling Android phone on the planet. The Galaxy A17 5G and A17 4G, the exact phones the A18 replaces, both made the list.

None of Samsung’s flagships cracked the top 10. Not the Galaxy S26 Ultra. Not the S26. The A-series is Samsung’s real volume engine — it’s where the company moves units by the millions across Africa, Latin America, Southeast Asia, and the Middle East. When Samsung updates the A18, it’s not chasing tech enthusiasts. It’s protecting market share in regions where a $200 phone is a major purchase.

That context also explains the pricing pressure. The global smartphone market is heading for a steep decline in 2026, driven by memory chip shortages caused by AI data center demand. Samsung’s already raised prices on its flagship Galaxy S26 and foldable lines. The A18 may not escape that squeeze — early estimates suggest prices could land slightly above the A17’s launch tags of roughly $200 (5G) and $170 (4G).

Should You Wait for the Galaxy A18?

If you need a phone today, the Galaxy A17 is still a strong buy — especially at current discounts. The A18 is a modest refresh, not a generational leap. The Snapdragon chip swap on the 5G model is the most meaningful upgrade, and even that won’t dramatically change daily performance for most users.

Mass production is reportedly starting in August, which means the A18 will likely arrive in the fall. If you can wait a few months, you’ll get a slightly newer chip and the A17’s price will drop further. But don’t hold off expecting a redesign. The A18 is built to do exactly what the A17 did — sell in massive numbers to people who want a reliable Samsung phone without paying flagship prices.

Samsung’s attention right now is on the six devices coming to Galaxy Unpacked on July 22 — foldables, smart glasses, and new watches. The A18 won’t get that spotlight. But in pure sales terms, it’ll likely outsell every single one of them.

FAQs

What display upgrade does the Samsung Galaxy A27 bring over the A-series? 

Samsung upgraded the Galaxy A27 to a punch-hole display, dropping the Infinity-U notch used in the A17 and A18. It also added DeX desktop mode support. The A27 is Samsung’s first budget phone below the A3x tier to get a modern display cutout, signaling features may eventually trickle down to the A1x lineup.

How are rising memory chip costs affecting phone prices in 2026? 

AI data centers are consuming massive amounts of high-bandwidth memory, driving up LPDDR5X prices by roughly 89% quarter-over-quarter. That’s adding an estimated 13% to average global smartphone prices, hitting budget and mid-range devices hardest — with no relief expected before late 2027.

What Samsung budget phones have the best battery life right now? 

The Galaxy M47 5G packs a 6,000mAh battery — the largest in Samsung’s current mid-range lineup. It also features bypass charging to reduce heat during gaming. For buyers prioritizing endurance, Samsung’s M-series battery specs currently outperform the A-series across the board.

How does the Galaxy A57 compare to the A56 for value? 

The A57 costs $50 more at launch ($549 vs. $499), while the A56 has dropped to around $385 at retailers. Performance gains are modest at roughly 12–15% CPU improvement. For most buyers, the A56 at its current price offers significantly better overall value.

Is Samsung upgrading cameras across its 2027 phone lineup? 

Leaks suggest the Galaxy S27 series will get a new 200MP sensor and ultrawide upgrade — Samsung’s first major camera hardware change in four years. The S27 Pro model may also carry a privacy display feature currently reserved for the S26 Ultra, signaling camera and display tech trickling down.

The post Samsung Galaxy A18 Leaked Worth The Wait With The New Chip appeared first on Memeburn.

The launch arrives on one of the busiest weeks for AI releases in 2026. OpenAI confirmed hours later that GPT-5.6 Sol, Terra, and Luna will become publicly available on Thursday, following weeks of restricted access at the request of the Trump administration.

What Grok 4.5 Does

SpaceXAI built Grok 4.5 as a general-purpose model targeting coding, agentic workflows, office productivity, and knowledge work. According to the company’s announcement, the model excels at software engineering tasks, end-to-end app building, and legal work, where it scored first on Harvey’s Legal Agent Benchmark.

The model is available immediately in Grok Build (SpaceXAI’s developer CLI), in Cursor across all subscription plans, and through the SpaceXAI API console. SpaceXAI also offers plugins for Microsoft Word, PowerPoint, and Excel, positioning Grok 4.5 as a tool for both developers and enterprise office workers.

One notable restriction is that Grok 4.5 is not available in the EU, with SpaceXAI targeting mid-July for European access. The company did not explain the delay, though regulatory compliance is the likely factor.

Benchmark Results Tell a Mixed Story

SpaceXAI published four benchmark scores at launch. The results show a model that is competitive with some leading alternatives but not the best performer in any category.

• DeepSWE 1.0 (each provider’s harness): Fable 5 at 66.1%, GPT 5.5 at 64.3%, Grok 4.5 at 62%, Opus 4.8 at 55.8%
• DeepSWE 1.1 (standardized harness): Fable 5 at 70%, GPT 5.5 at 67%, Opus 4.8 at 59%, Grok 4.5 at 53%
• Terminal Bench 2.1: Fable 5 at 84.3%, GPT 5.5 at 83.4%, Grok 4.5 at 83.3%, Opus 4.8 at 78.9%
• SWE Bench Pro: Fable 5 at 80.4%, Opus 4.8 at 69.2%, Grok 4.5 at 64.7%, GPT 5.5 at 58.6%

Grok 4.5 performs strongest on Terminal Bench 2.1, where it nearly matches GPT 5.5 and trails Anthropic’s Fable 5 by just one percentage point. On DeepSWE 1.1, the gap widens considerably, with Grok 4.5 scoring 17 points behind the leader.

Two important caveats apply. First, all benchmark figures were reported by SpaceXAI using vendor-provided harnesses and competitor scores drawn from published system cards. Independent verification is still pending. Second, the benchmarks compare Grok 4.5 against GPT 5.5, not GPT 5.6, which launched the same day but after Grok 4.5’s announcement.

Where the Real Advantage Shows Up

Musk’s pitch for Grok 4.5 is not about benchmark leadership. It is about cost-per-task efficiency.

SpaceXAI claims Grok 4.5 uses 4.2 times fewer output tokens than Opus 4.8 when completing SWE Bench Pro tasks. According to SpaceXAI’s data, Grok 4.5 averaged 15,954 output tokens per task compared to 67,020 for Opus 4.8. Combined with lower per-token pricing, this means a developer running high-volume coding workflows could spend a fraction of what they would on Anthropic’s or OpenAI’s flagships.

The pricing comparison across frontier models makes the gap clear:

• Grok 4.5: $2 input / $6 output per million tokens
• Claude Opus 4.8: $5 input / $25 output
• Claude Fable 5: $10 input / $50 output
• GPT 5.6 Sol: $5 input / $30 output
• GPT 5.6 Luna: $1 input / $6 output

OpenAI’s Luna tier matches Grok 4.5’s output rate and undercuts it on input, though Luna is a lightweight model designed for speed and volume rather than frontier reasoning. At the premium end, Grok 4.5 is roughly 60% cheaper on input and 76% cheaper on output than Opus 4.8.

For teams that already struggle with token costs, the math is straightforward. If Grok 4.5 can deliver Opus 4.7-level quality at these prices and with fewer tokens per task, it becomes a strong candidate for production workloads where raw benchmark supremacy matters less than cost-per-finished-task.

The Cursor Factor

Grok 4.5 is the first model built jointly by SpaceXAI and Cursor following their partnership announcement in April. Cursor CEO Michael Truell’s team contributed trillions of tokens of developer session data, including real code edits, debugging traces, and user-agent interactions from the Cursor platform.

This is a significant training data advantage that none of SpaceXAI’s competitors have access to. Cursor crossed $1 billion in annualized revenue in November 2025 and serves a large base of professional software engineers. The real-world coding data generated by those users is fundamentally different from static code repositories. It captures how developers actually debug, iterate, and build software with AI assistance.

The training ran on tens of thousands of Nvidia GB300 GPUs inside SpaceXAI’s Colossus data center in Memphis. Axios reported an unusual detail about this infrastructure: SpaceXAI is currently leasing some of the same compute capacity to its competitors Anthropic and Google. As SpaceXAI’s own model ambitions grow, it may eventually face a decision between using that capacity for internal training or continuing to collect revenue from rivals.

That tension could become a strategic inflection point. Right now, the leasing revenue supports SpaceXAI’s business case for investors. But every GPU hour spent training Grok is a GPU hour not generating cloud revenue. How SpaceXAI manages that balance will likely shape both its AI competitiveness and its financial performance as a public company.

A Crowded Week for AI Launches

The timing of Grok 4.5’s release is not coincidental. OpenAI confirmed on Tuesday evening that GPT-5.6 Sol, Terra, and Luna will launch publicly on Thursday, July 9. The models had been restricted to roughly 20 government-approved partner organizations since June 26, following a White House request tied to cybersecurity concerns.

Musk’s decision to release Grok 4.5 one day before GPT-5.6 goes public ensures his model captures attention before OpenAI dominates the news cycle. It also means Grok 4.5’s benchmarks avoid a direct comparison to GPT 5.6 Sol, which OpenAI says achieves 91.9% on Terminal Bench 2.1, substantially higher than Grok 4.5’s 83.3%.

Musk himself was transparent about where Grok 4.5 sits in the competitive landscape. In a post on X, he wrote that SpaceXAI’s internal assessment puts Grok 4.5 as “roughly comparable to Opus 4.7, but much faster.” Opus 4.7 is Anthropic’s previous flagship, already superseded by Opus 4.8 in May. Comparing Grok 4.5 to a model that is one generation behind Anthropic’s current best is tactically honest, but it also reveals the gap SpaceXAI still needs to close at the top end of the benchmark leaderboard.

The Risk SpaceXAI Is Actually Betting On

The implicit bet behind Grok 4.5 is that the AI market is splitting into two segments, and most revenue will flow to the cheaper one.

One segment is the frontier race, where labs compete for the highest possible benchmark scores on the hardest tasks. That segment matters for prestige, research credibility, and a small number of enterprise use cases that genuinely require the most powerful models. Fable 5, GPT 5.6 Sol, and Opus 4.8 compete here.

The other segment is production-grade AI at scale, where the question is not “which model scored highest on SWE Bench Pro” but “which model finishes my task at an acceptable quality for the lowest cost.” This is where Grok 4.5 is positioned. At $2/$6 per million tokens and with roughly one-quarter the token usage of Opus 4.8 per task, Grok 4.5 could deliver completed coding tasks at a fraction of the cost, even if individual benchmark scores trail the leaders.

This mirrors what Chinese AI vendors like DeepSeek and Zhipu have been doing: getting close enough on performance and winning on price. The difference is that SpaceXAI has a distribution channel that Chinese vendors lack in Western markets, through Cursor’s developer user base, the X platform, and enterprise plugins for Microsoft Office.

Whether developers actually switch will depend on whether the token efficiency claims hold in practice and whether Cursor’s model-agnostic users accept Grok as a default. Cursor currently lets developers choose between Claude, GPT, and Gemini. If SpaceXAI pushes Grok 4.5 too aggressively as the default, it risks alienating the power users who made Cursor valuable in the first place.

What to Watch Next

Grok 4.5 arrives at a moment when all three major Western AI labs are releasing significant models simultaneously. GPT 5.6 Sol goes public on Thursday with reported benchmark scores that substantially exceed Grok 4.5 on several key tests. Anthropic’s Fable 5 has returned to service after 19 days offline under a U.S. export control directive. And SpaceXAI is trying to carve a market position defined by cost rather than capability leadership.

The next test for Grok 4.5 will be independent benchmarking. SpaceXAI’s self-reported numbers paint a competitive picture, but third-party validation, particularly on the token efficiency claims, will determine whether the model’s value proposition holds beyond the launch day marketing.

SpaceXAI’s Cursor acquisition is expected to close in Q3 2026. How the combined company handles model selection inside Cursor after that closing will be a signal of whether SpaceXAI prioritizes ecosystem openness or internal model adoption.

FAQs

How much does Grok 4.5 cost compared to Claude Opus 4.8?

Grok 4.5 is priced at $2 per million input tokens and $6 per million output tokens. Claude Opus 4.8 costs $5 input and $25 output, making Grok 4.5 roughly 60% cheaper on input and 76% cheaper on output per million tokens.

Is Grok 4.5 better than GPT 5.6 Sol?

Grok 4.5 and GPT 5.6 Sol launched within hours of each other. SpaceXAI’s benchmarks compare Grok 4.5 to GPT 5.5, not 5.6. Early reports indicate GPT 5.6 Sol scores significantly higher on Terminal Bench 2.1, but Grok 4.5 is substantially cheaper per token.

Can I use Grok 4.5 in Europe?

Not yet. SpaceXAI says Grok 4.5 is not available in the EU at launch, with European access expected in mid-July 2026. The company has not explained the specific reasons for the delay.

What is Grok 4.5’s token efficiency advantage?

SpaceXAI claims Grok 4.5 uses 4.2 times fewer output tokens than Opus 4.8 on SWE Bench Pro tasks, averaging 15,954 tokens per task versus 67,020. If confirmed by independent testing, this would significantly reduce the effective cost of running coding workloads.

The post Grok 4.5 Launches With Budget Pricing and Cursor Integration appeared first on Memeburn.

Huawei Pura 90s Gets a New Name — and a New Look

The Pura 90 series launched in China back in April 2026, but for international markets, Huawei’s tacking on an “S” — calling it the Pura 90s. The company hasn’t explained what that letter means. It could hint at minor spec tweaks for global hardware or simply serve as a branding distinction between regions.

Energy, blended. Vibes, refreshed. #HUAWEIPura90s Pro – Guava Soda, Orange Soda, Coconut White, and Mulberry Black.🥤#NowIsYourMoment #HuaweiLaunch pic.twitter.com/xS2iF8rdRO

— Huawei Mobile (@HuaweiMobile) July 8, 2026

What’s clear is the design direction. The Pura 90s Pro arrives in four colors — Guava Soda, Orange Soda, Coconut White, and Mulberry Black. The first two use an angular gradient that sweeps color diagonally across the back panel, blending shades at the center. It’s a deliberate departure from the muted tones Huawei used on recent flagships.

This is the most visually aggressive Pura phone since the P30 Pro. Huawei’s clearly chasing a younger demographic here, and the fruity naming convention tells you exactly who they want holding this phone. Whether that alienates fans who liked the understated premium look of the Pura 70 and Mate 60 lines is a real question — but it’s a bet Huawei seems willing to make.

The 200MP Telephoto Is the Real Headline

Gradient colors get the Instagram clicks, but the Pro Max’s camera system is where things get serious. It carries a 200MP periscope telephoto lens with a 1/1.28-inch sensor — the largest ever put in a smartphone telephoto module. That’s paired with 4x optical zoom, RYYB color filtering, and CIPA 7.0-level image stabilization.

The main camera is a 50MP RYYB sensor with LOFIC technology (Lateral Overflow Integration Capacitor) — a feature borrowed from Sony’s high-end imaging division that improves dynamic range by preventing pixel overflow. It supports a 10-stop variable aperture from f/1.4 to f/4.0. The third lens is a 40MP ultrawide at f/2.2.

Both the Pro and Pro Max use Huawei’s XMAGE imaging pipeline, which handles post-processing with a focus on balanced color output rather than heavy saturation.

For context, Samsung’s upcoming Galaxy S27 series is also pushing camera upgrades this cycle — but a 200MP telephoto at this sensor size puts Huawei in a category of its own for now. The real test will come down to software processing, and that’s where Huawei’s lack of Google-backed computational photography tools could be a limiting factor.

Kirin 9030S and HarmonyOS 6.1 — Power Without Google

Under the hood, both the Pro and Pro Max run the Kirin 9030S chipset with up to 16GB of RAM and 1TB of storage. The standard Pura 90 (if it joins the global launch) would get the lower-tier Kirin 9010S.

The software story is HarmonyOS 6.1, which brings an agentic AI assistant and a refreshed set of image editing tools. But here’s the trade-off that matters most to global buyers: no Google Play, no Gmail, no YouTube, no Maps — at least not natively. Huawei’s AppGallery is the default storefront.

That’s been Huawei’s reality since 2019, and it remains the single biggest barrier for anyone outside China. If you’re considering this phone, the camera and design need to be compelling enough to justify living inside Huawei’s ecosystem — or comfortable enough sideloading apps to fill the gaps.

The 2026 flagship phone market is stacked right now, from Samsung’s foldables to Google’s Pixel 11 push. Huawei’s hardware competes on paper. Its software ecosystem still doesn’t, and that’s a trade-off you shouldn’t ignore.

Global Pricing and Pre-Order Details

Huawei will reveal official global pricing on July 14 at the Kuala Lumpur launch event. In China, the base Pura 90 starts around $690, while the Pro Max pushes past $1,100 depending on storage configuration.

In Malaysia, pre-orders are already live. A RM100 deposit unlocks rebates of up to RM300 for the Pura 90s Pro and up to RM400 for the Pro Max. Whether the standard Pura 90 (or 90s) will be part of the global roster is still unconfirmed — Huawei’s teasers have only shown the Pro and Pro Max so far.

It’s worth watching how Huawei prices these against the Galaxy Z Fold 8 and Pixel 11 Pro — two devices that arrive in the same window. If Huawei can undercut Samsung’s ultra-premium tier while delivering this camera hardware, the Pura 90s Pro Max becomes a genuinely interesting proposition for photography-focused buyers willing to forgo Google services.

Should You Care About the Pura 90s?

Honestly? It depends on what you value most in a phone. If camera hardware is your top priority and you’re open to life outside Google’s walled garden, the Pura 90s Pro Max packs specs that no competitor currently matches on the telephoto end. The gradient design is also a refreshing swing — reminiscent of how Nothing’s color-focused editions have turned aesthetics into a marketing advantage.

But if app compatibility, Google services, and ecosystem interoperability are non-negotiable, the HarmonyOS trade-off is still too steep for most global users. Huawei knows this. The Pura 90s isn’t trying to win everyone — it’s trying to win a specific slice of the market that cares more about imaging and design than app store breadth.

We’ll have a clearer picture after July 14. Until then, the teasers are doing exactly what Huawei wants: getting people talking about a phone brand most of the world had started to forget.

FAQs

What is HarmonyOS and how does it differ from Android? 

HarmonyOS is Huawei’s proprietary operating system, developed after the company lost access to Google services in 2019. Unlike Android, it doesn’t include Google Play or core apps like Gmail or Maps by default. Instead, it uses Huawei’s AppGallery as its primary app store. HarmonyOS is designed to work across phones, tablets, wearables, and IoT devices using a shared microkernel architecture.

How does smartphone camera sensor size affect photo quality? 

A larger sensor captures more light, which directly improves detail, dynamic range, and low-light performance. That’s why the Pura 90s Pro Max’s 1/1.28-inch telephoto sensor matters — most smartphone camera systems use far smaller telephoto sensors, which limits zoom quality in dim conditions.

What are gradient colors on smartphones? 

Gradient coloring blends two or more colors across a phone’s back panel, creating a smooth transition effect. Huawei popularized this with the P30 Pro in 2019, and several brands, including Nothing with its limited editions, now use gradient or color-focused designs to stand out in a crowded market.

Can you install Google apps on Huawei phones? 

Not officially. Since Huawei lost its Google license, its phones ship without Google Mobile Services. Some users sideload Google apps using third-party methods, but it’s not guaranteed to work reliably — especially with apps that depend on Google Play Services for notifications, payments, or authentication.

What is XMAGE in Huawei phones? 

XMAGE is Huawei’s proprietary imaging brand, replacing its former Leica partnership. It covers both hardware (lens design, sensor tuning) and software (post-processing algorithms). The system prioritizes natural color reproduction over the heavy contrast and saturation some competitors favor, aiming for more true-to-life photos.

The post Huawei Pura 90s Goes Global With 200MP Cam, Bold 2026 appeared first on Memeburn.