OpenAI's biggest pre-train 'Doug' surfaces ahead of GPT-6

Share
OpenAI's biggest pre-train 'Doug' surfaces ahead of GPT-6

OpenAI is reportedly gearing up to restart base-model scaling after two years of squeezing capability out of post-training — while Apple quietly tests Chinese memory chips to survive the AI-driven supply crunch.

OpenAI is reportedly pre-training its largest model ever, a project codenamed "Doug" that could land before the end of the year — and it is not GPT-6. ChrisGPT, an X user who has tracked OpenAI's model pipeline for years, said on August 9 that Doug is the biggest pre-training run OpenAI has attempted to date and is separate from GPT-6, which he believes is "Astra" — the flagship model OpenAI abruptly paused last week over security concerns. Doug's expected no later than November, he said. The claim isn't fresh from one leaker: SemiAnalysis, in a research memo dated July 9 that it republished this week, wrote that OpenAI has overcome its pre-training challenges and is "actively advancing" a much larger model under the Doug codename.

If accurate, Doug marks a strategic turn. For nearly two years OpenAI's capability gains have come mostly from post-training, reinforcement learning and inference-time compute layered on aging base models — the approach that made the Astra pause sting less, since the frontier kept moving without a new foundation model. A Doug-class pre-train would signal OpenAI believes the raw scaling laws still have room, and that fixing its earlier pre-training bottlenecks was worth the enormous compute bill. Read it together with the Astra pause — we covered that last week — and the picture is of a lab managing two frontiers at once: shipping safety-verified models while rebuilding its base-model muscle. OpenAI hasn't commented, and skeptics note there's no release plan or API attached to the rumor yet, so treat the November timeline as a leak, not a roadmap.


Apple is testing memory chips from China's CXMT across iPhones and MacBooks, the Wall Street Journal reported Sunday, citing people familiar with the matter — a bid to ease an AI-driven supply crunch that's been pushing up prices. Apple has held early talks with ChangXin Memory Technologies about supplying components, with an eye to using them in some devices sold in China, and hopes to win White House approval for any deal. The stakes are political: U.S. rules bar American companies from sharing technology with CXMT, which effectively rules out the custom-spec chips Apple typically orders. HP and Acer have already started using CXMT memory in devices sold outside the U.S., and CXMT — now mainland China's most valuable listed company after its Shanghai listing — is the strongest challenger yet to Samsung, SK hynix and Micron's grip on DRAM. Apple has been raising device prices worldwide, blaming surging memory costs as AI companies hoover up capacity, so a CXMT route would be both a cost lever and a geopolitical statement — which is exactly why it's moving slowly. Chinese tech press frames it as part of a broader opening, with Baidu and Qwen models also edging into Apple's ecosystem.

What to watch: whether OpenAI confirms Doug by year's end, and whether the White House lets Apple's CXMT talks go anywhere.

If OpenAI is quietly restarting base-model scaling, does the Astra pause look more like triage than retreat? Write to us via our contact page.

Source: WinCentral

Get SiliconNoon in your inbox

One email a day with the AI stories that matter — what happened, the numbers, and why it matters. No spam, unsubscribe in one click.

Read more

OpenAI ships GPT-5.6-Cyber to vetted defenders under Daybreak

OpenAI ships GPT-5.6-Cyber to vetted defenders under Daybreak

Three days after OpenAI said it could not rule out "Critical" cyber capabilities in its next model, it is doing the opposite of slamming the door: shipping a more permissive cyber model to a vetted circle of defenders — and framing it as the only way to close a narrowing defense window. OpenAI released GPT-5.6-Cyber, a cybersecurity-tuned variant of GPT-5.6 Sol, and expanded its Daybreak program with two access tiers that put frontier cyber models in approved defenders' hands.

Attack exposes hidden reasoning traces inside Claude, GPT, Gemini

Attack exposes hidden reasoning traces inside Claude, GPT, Gemini

Two stories frame the day: researchers found a way to pry open the hidden chain-of-thought inside frontier models, and Germany moved to treat Meta's AI glasses the way it treated a spy doll. Researchers from the University of Tübingen, the Max Planck Institute, MATS Research, and security firm Snyk have found a way to extract the hidden reasoning traces of frontier models — and the method turned up the strongest evidence yet that some Chinese open-weight models were trained on US rivals'…

Upstage ships Solar Pro 4, an agentic model with 512K context

Upstage ships Solar Pro 4, an agentic model with 512K context

Korean AI lab Upstage rounded out its 2026 lineup this week with Solar Pro 4, a closed-API flagship aimed squarely at document-heavy agent work, roughly a week after the open-weights Solar Open 2. It is the company's clearest bet yet that the next AI battleground is finishing office jobs, not passing chat tests. Upstage released Solar Pro 4, a proprietary agentic model for multi-document, multi-tool workloads, with a 512K-token context window and up to 128K output tokens.

OpenAI previews Ultrafast mode for GPT-5.6 Sol

OpenAI previews Ultrafast mode for GPT-5.6 Sol

The speed wars just got a new front: OpenAI put its smartest model in a fast lane, Google shipped another Flash on a three-week cadence, and DeepSeek open-sourced the harness layer of its agent stack. Three moves, one theme — nobody is waiting. OpenAI is previewing Ultrafast, a new service tier that runs GPT-5.6 Sol up to 14 times faster than standard processing, generating as many as 750 output tokens per second on Cerebras hardware — with no quality compromise, the company says.

Mimir v1: 1B model trained on permissible data matches 4B rivals

Mimir v1: 1B model trained on permissible data matches 4B rivals

Two research stories this morning, one about what open models are trained on, one about what self-improving agents remember. Denmark's Foundation Models project says a 1-billion-parameter model built only on legally clean data goes toe-to-toe with models four times its size, and a new study quantifies how easily agents bake unsafe behavior into their persistent skills. Danish researchers say their 1B model, trained only on permissible data, matches rivals four times its size.

Fine-tuning CLIP's image tower wrecks out-of-distribution accuracy

Fine-tuning CLIP's image tower wrecks out-of-distribution accuracy

An IJCAI 2026 paper from Zhejiang University and Swansea University delivers the kind of result that quietly changes how engineers ship vision-language models: the branch you've been fine-tuning hardest is the one destroying your production accuracy. A team led by Zhejiang University's Jingyuan Chen found that aggressively adapting CLIP's image encoder — the standard reflex when customizing a vision-language model — actively destroys its ability to handle out-of-distribution (OOD) data.