China has set an official deadline for the solid-state battery era, but even the country's biggest battery maker is openly sceptical it can be met. A five-year plan for the battery industry, dated September 14 and issued by the Ministry of Industry and Information Technology with six other agencies, commits China to "initial large-scale application" of all-solid-state batteries by 2030, according to CnEVPost. To put this into context, even the most ambitious forecasts have not committed to this kind of roll-out speed.

The document is unusually specific. It demands that long-life lithium batteries reach 15,000 charge-discharge cycles and that leading manufacturers cut defect rates to parts-per-billion levels, and it calls for implementation of  the isostatic presses solid-state production needs.

The measured response comes from CATL. Chairman Robin Zeng rates the technology four out of nine on a readiness scale, and sees little chance of it reaching a million vehicles before 2030. CATL plans small-batch production in 2027. Zeng's caution carries weight: CATL runs the world's first 5 GWh all-solid-state line, in Hefei.

The stakes reach beyond China. Taiwan's ProLogium broke ground in February on a plant at Dunkirk, France, beside the Gravelines nuclear station — 0.8 GWh by 2028, 12 GWh by 2032.

Beijing has supplied the timeline. Industry, for now, supplies the realism. continue reading

The post China Sets 2030 Deadline for Solid-State Battery Tech, as CATL Says They’re Not Ready Yet appeared first on The Red Ferret Journal.

Quick Brief

Featured story today: "Classified estimates show the NSA is paying billions to test AI models" And more models & releases news from the AI universe –

continue reading

The post AI News – October 08, 2026 appeared first on The Red Ferret Journal.

Quick Brief

Featured story today: "Classified estimates show the NSA is paying billions to test AI models" And more models & releases news from the AI universe –

continue reading

The post AI News – September 25, 2026 appeared first on The Red Ferret Journal.

Tesla has ramped Optimus production at its Fremont factory to several hundred humanoid robots a week — roughly ten times its second-quarter output — but the robots still cannot handle general-purpose work, according to Electrek.

The Information reported that Tesla went from a few dozen Optimus units a week during Q2 testing to several hundred a week in August. Managers want a continuous automated line making more than 1,000 robots a week by year-end, with an eventual target of about 20,000 a week. The line occupies the space where the Model S and Model X were built, after Tesla ended production of both in early May.

The bottlenecks are physical. Each hand and forearm contains over 100 screws and small components that workers still assemble by hand, and fixtures cannot consistently align parts far tighter-tolerance than car parts, producing more robots needing rework. Three people familiar with the system told The Information that Optimus' AI cannot yet reliably handle a wide range of tasks, and that learning a basic task still takes several days.

Most units are used internally for testing, training and data collection. Elon Musk said in January 2025 that Tesla would build about 10,000 Optimus robots that year, with "several thousand" doing useful work; he later admitted zero were.

Competitors are not waiting. XPeng started an automated IRON humanoid production line in Guangzhou this month, targeting commercial sales in 2027.

Photo by Matthew Pablico on Unsplash. continue reading

The post Tesla Ramps Optimus to Hundreds of Robots a Week, But They Still Can’t Handle General Tasks appeared first on The Red Ferret Journal.

All-solid-state batteries have crossed a line the industry has been promising for years, mass production. Taiwan's ProLogium Technology announced on September 2 that its Generation 3.5 Lithium Ceramic Battery (LCB) has entered mass production at its "Giga-level" facility in Taoyuan, according to CarNewsChina.

ProLogium's 185.4 Ah large-format cell was measured at 381 Wh/kg gravimetric and 903 Wh/L volumetric energy density, with the energy-density figures verified by TÜV Rheinland. Separately, UL Solutions tested the cell under China's GB/T 43568-2026 methodology, holding it for six hours under continuous vacuum at 120°C; ProLogium recorded less than 0.05% mass loss, well under the standard's 0.5% threshold, which qualifies the cell as all-solid-state. The company has a technology cooperation agreement with Mercedes-Benz, and says cumulative LCB shipments have passed 2.4 million cells.

The caveats matter as much as the milestone. Taoyuan's initial operational capacity is just 0.5 GWh, rising toward 1-2 GWh — theoretically about 6,250 packs of 80 kWh a year, and far short of the volume mainstream passenger EVs need. And 381 Wh/kg is a cell-level figure: at 0.5 GWh, the harder test is whether ProLogium can reach competitive cost, yield and pack-level performance. It is aiming to scale in Dunkirk, France, eventually.

Once again it's proving easier to announce stuff than actually build it when it comes to solid state batteries. But if we believe all the recent statements, 2027 is likely to see a flood of small scale production across the globe, particularly in China.

Photo by Oxana Melis on Unsplash continue reading

The post Solid-State Batteries Finally Reach Mass Production — But the Factory Is Tiny appeared first on The Red Ferret Journal.

Xpeng has commissioned an automated production line for its Iron humanoid robot, and the first machine assembled on it walked off the line by itself. The Chinese carmaker announced the milestone on Tuesday, saying the event marks Iron's move from research and development prototypes to production-line manufacturing.

The line was designed and built by Xpeng's own in-house development team, according to CnEVPost, and more than 80 percent of its core processes are automated. Chairman and chief executive He Xiaopeng said the production lines "were created from scratch with no precedent to follow". Xpeng calls the plant the world's first automated line for advanced humanoid robots, but has not disclosed its capacity or the robot's price.

Iron has 76 degrees of freedom, 21 in each hand, and runs on three in-house Turing AI chips delivering up to 2,250 TOPS, enough for Xpeng's physical AI model to run on board without remote control. While commissioning is not the same as mass production, manufacturing in volume is targeted for the end of 2026, with an official launch and deliveries in China and overseas in 2027, starting at XPeng's own stores and campuses.

Some of you will remember the Iron as the one which featured such a seductive walk that people accused the company of hiding a human operator inside the shell. In one of the launch events, Xpeng had to cut open the leg of the robot to prove it was made up of pulleys and levers. Don't panic, but the androids are clearly almost here.

Photo Xpeng continue reading

The post Xpeng’s Iron Humanoid Robot Walks Off Its Own Production Line appeared first on The Red Ferret Journal.

OpenAI has given major customers a first look at Astra, an upcoming frontier AI model family, at a closed-door preview in San Francisco, TIME reports.

In one demonstration, 16 AI agents divided a research-level mathematics problem into subproblems, coordinated their work and assembled a proposed proof. In another, Astra navigated well-known desktop software, creating and editing work across applications "with unnerving speed." CEO Sam Altman told the visitors that watching Astra use a computer in a "super-human, very fast kind of way" had been among the most striking moments for employees.

Altman explained that Astra would enable "persistent agents" — virtual colleagues toiling for sustained periods on tasks. Its biggest impact, he predicted, would come from people using it to discover new knowledge. "I expect this will be the first model where the model actually invents new things in a way that matters," Altman told the group. "That's a very AGI-like thing."

The preview comes amid turmoil: OpenAI recently acknowledged that an unreleased agent escaped its sandbox during testing and hacked into production systems at Hugging Face to obtain benchmark answers, and the company has since frozen some experiments, paused a major training run and tightened its safeguards. TIME reports OpenAI still plans to ship Astra, but its release now depends on clearing those new safety measures.

For more information, see also this Testing Catalog report.

Photo by Aerps.com on Unsplash. continue reading

The post OpenAI Shows Off Astra: 16 AI Agents, ‘Super-Human’ Desktop Control appeared first on The Red Ferret Journal.

Alibaba recently released Qwen3.8-27B, an open AI model designed to run on consumer hardware like a laptop, and open-sourced the weights of its most powerful model, Qwen3.8 Max. It was a direct answer to Meta's return to open, locally-run AI.

The Chinese tech giant, which until recently has had to play second fiddle to the mighty DeepSeek hype, claims that Qwen3.8-27B has "excellent capabilities" in coding, professional work, research and long-horizon agentic tasks, matching the performance of a model ten times its size. Running a model on-device — on a laptop or phone — can be faster and more secure, industry experts say, because data stays on local hardware instead of being sent to cloud data centres.

Guess what? The company isn't lying. Qwen is an absolute beast of a AI model. Not just because it's free and open source, but because you can run it on extremely modest hardware and get super impressive results out of it. I've been playing with it over the past couple of weeks and I'm blown away by just how performative this model is.

Anything from creating fun Android games to producing full blown websites and mobile applications. Sure, it takes longer versus the big cloud models like Claude, but producing something in an hour versus fifteen minutes is neither here nor there when you consider that you're not spending a penny on a subscription of any kind.

"The company which can offer the most capable open weights models will move ahead in this race," Neil Shah of Counterpoint Research told CNBC. With capable models that fit in a laptop now topping Hugging Face's trending charts, the race to run AI on our own machines is heating up.

Photo by Daniil Komov on Unsplash. continue reading

The post Alibaba’s new Qwen3.8-27B: A game changing AI Model That Fits on a Laptop appeared first on The Red Ferret Journal.

Biotech pioneer Moderna and drugmaker Merck have announced a personalized messenger-RNA cancer therapy, designed to tamp down the recurrence and spread of high-risk melanoma in a late-stage trial – the first successful Phase 3 result ever for an mRNA cancer treatment. According to the companies’ announcement, the INTerpath-001 study enrolled 1,137 patients to receive the made-to-order vaccine Intismeran (V940, or mRNA-4157), alone or alongside the Keytruda antibody treatment.

The combination met its primary endpoint, recurrence-free survival, and a key secondary endpoint, distant metastasis-free survival. The full report is due shortly. Intismeran is built from a patient’s tumor: scientists sequence it to map the cancer’s unique mutational fingerprint and craft an mRNA strand targeting up to 34 neoantigens to train the immune system to attack the disease. An interesting example of using fire to fight fire.

Moderna CEO Stéphane Bancel called the findings a pivotal moment for cancer research; study lead Georgina Long, medical director of the Melanoma Institute Australia, called them a landmark moment for adjuvant melanoma treatment; and Merck Research Laboratories president Dean Y. Li said they reinforce the promise of a more personalized approach. The companies said they will seek regulator approval to make an appropriate filing, and to discuss the potential to test the same treatment protocol against lung, kidney, bladder and other cancers. The fight continues!

Photo by Ian Hutchinson on Unsplash. continue reading

The post Moderna and Merck’s Personalized mRNA Cancer Vaccine Succeeds in Landmark Phase 3 Melanoma Trial appeared first on The Red Ferret Journal.

The tropical Pacific is warming so quickly that this year’s El Niño could become the strongest ever recorded, China’s National Climate Center warned this week, according to China Daily. Sea surface temperatures in the central and eastern equatorial Pacific have risen sharply since El Niño conditions emerged in May, with the three-month running average reaching 2.09 C above normal in July, up 0.49 C from June, and climbing to 2.64 C in the first half of August.

At the current pace, forecasters expect that running average to exceed 2.5 C — the threshold for an extreme “super” El Niño event. The center says temperatures are likely to peak around November or December, potentially surpassing the three strongest episodes on record — 1982-83, 1997-98 and 2015-16.

China is already feeling developing-phase effects: rainfall concentrated in two major belts across the north and south, a stronger summer monsoon lifting typhoon activity, and overlapping rain systems heightening the risk of compound disasters. Because the climate system responds to El Niño with a time lag, effects could linger into autumn — more rain across parts of eastern, southern and southwestern China while much of the country runs warmer than average. The center says it will strengthen monitoring and forecasting as the event develops.

Photo by Matt Palmer on Unsplash. continue reading

The post El Niño Could Be the Strongest on Record, China’s Climate Center Warns appeared first on The Red Ferret Journal.

Memory-chip stocks are stalling as price growth cools, with analysts warning that the explosive AI-driven memory-price boom may be peaking, according to the South China Morning Post.

Morgan Stanley said in a report last week that the memory cycle is set to enter its late stage in the fourth quarter, as price increases moderate and inventories build. Shares of the biggest makers — Micron, SK Hynix and SanDisk — have retreated from recent highs. Bernstein Research said conventional DRAM contract prices were expected to rise about 17% in the third quarter, down from the roughly 65% jump in the April–June period.

Chinese producers are adding supply that could temper the cycle. UBS expects ChangXin Memory Technologies (CXMT) to nearly double its monthly DRAM capacity, from around 240,000 wafer starts to 466,000 by late 2028, lifting its share of global DRAM bit supply from about 7% to 10%. Yangtze Memory Technologies (YMTC) is expanding its NAND flash footprint.

The slowdown is uneven: PC and smartphone makers are resisting higher prices, while servers and AI data centers stay resilient and Citi Research, citing TrendForce, sees servers taking 51.1% of global NAND production by 2027, up from 44.2% in 2026. As Morgan Stanley put it, “a strong AI cycle and a cyclical downturn can coexist,” though JPMorgan still expects shortages to persist for another two years.

Photo by J K on Unsplash. continue reading

The post AI Memory-Price Boom Shows Signs of Peaking, Analysts Warn appeared first on The Red Ferret Journal.

Quick Brief

Featured story today: "Qwen 3.8 27B is excellent, but it defaults to overthinking things" And more models & releases news from the AI universe –

continue reading

The post AI News – August 17, 2026 appeared first on The Red Ferret Journal.

Quick Brief

Featured story today: "Kimi K3 full model running on 16x GB10 cluster at 20+tps" And more models & releases news from the AI universe –

continue reading

The post AI News – August 5, 2026 appeared first on The Red Ferret Journal.

Quick Brief

Featured story today: "DeepSeek-V4-Flash Update" And more models & releases news from the AI universe –

continue reading

The post AI News – August 1, 2026 appeared first on The Red Ferret Journal.

Quick Brief

Featured AI story today: "Show HN: Bento - An entire PowerPoint in one HTML file (edit+view+data+collab)" And more models & releases news from the AI universe –

continue reading

The post AI News – July 23, 2026 appeared first on The Red Ferret Journal.

Quick Brief

Featured story today: "Testing llama.cpp MTP support on Qwen3.6 - RTX 5090" And more models & releases news from the AI universe –

continue reading

The post AI News – May 17, 2026 appeared first on The Red Ferret Journal.

Quick Brief

Featured story today: "Multi-Token Prediction (MTP) for LLaMA.cpp - Gemma 4 speedup by 40%" And more models & releases news from the AI universe –

continue reading

The post AI News – May 8, 2026 appeared first on The Red Ferret Journal.

Quick Brief

Featured story today: "Llama.cpp MTP support now in beta!" And more models & releases news from the AI universe –

continue reading

The post AI News – May 4, 2026 appeared first on The Red Ferret Journal.

Quick Brief

Featured story today: "I'm done with using local LLMs for coding" And more models & releases news from the AI universe –

continue reading

The post AI News – April 28, 2026 appeared first on The Red Ferret Journal.

Quick Brief

Featured story today: "I'm glad we have deepseek" And more models & releases news from the AI universe –

continue reading

The post AI News – April 25, 2026 appeared first on The Red Ferret Journal.

Quick Brief

Featured story today: "Kimi K2.6 Released (huggingface)" And more models & releases news from the AI universe –

continue reading

The post AI News – April 20, 2026 appeared first on The Red Ferret Journal.