White House misses AI review deadline + Alibaba ships Qwen3.8-Max

Executive Order 14409, signed June 2, gave federal agencies 60 days to stand up a classified benchmarking process, a voluntary framework for disclosing frontier models before release, and a plan to widen federal cybersecurity hiring. That clock ran out on August 1 with no Federal Register notice, nothing published by the National Institute of Standards and Technology or the Cybersecurity and Infrastructure Security Agency, and no statement from the Office of Science and Technology Policy. The framework would have given agencies a 30-day look at covered models before release — until it exists, labs are guessing which of their systems would even qualify.

Qwen3.8-Max is a mixture-of-experts model with 2.4 trillion total parameters, roughly 95 billion of them active per request, and a context window of about one million tokens. Alibaba reports 86.6 on Terminal-Bench 2.1 — ahead of Claude Opus 4.8 and Claude Fable 5 at 84.6, behind GPT-5.6 Sol at 88.8 — while trailing Fable 5 badly on SWE-bench Pro, at 67.7 against 80.0. The hosted API is live now at $2 per million input tokens and $6 per million output tokens. Open weights for Max and the smaller Qwen3.8-27B are promised for next week and are not published yet.

MiniMax launched H3 on July 31 and put the weights on Hugging Face on August 2, making it one of the few downloadable models that generates video with matching audio rather than silent clips. The 33 billion parameter system reads text, images, video and audio in a single context and returns 4 to 15 seconds at up to 2K resolution, 24 frames per second, with 32 kilohertz stereo sound. The MiniMax Community License is narrower than "open" suggests: it bars use of the weights outright in the United States, European Union, United Kingdom and South Korea, requires written authorization above $20 million in revenue, and mandates on-screen attribution. The hosted API is unrestricted.

Jacob Tsimerman, who won this year's Fields Medal for his part in proving the André-Oort conjecture, used a press conference after the ceremony to announce he is leaving the University of Toronto for OpenAI. He is only the second Canadian to take the prize, which is often described as mathematics' highest honor. Tsimerman co-authored a paper last year on futures in which AI causes mass human death, and has said that AI safety is the most important question of our time.

In a new essay, the veteran engineer Steve Yegge argues that teams running coding agents at real scale have to abandon both continuous integration and human approval gates, because neither survives the commit volume. He reports averaging 175 real commits a day with peaks near 250, and batching 120 to 150 of them at once instead of merging in sequence. His own July usage ran to 69 billion tokens, 96 percent of which were served from cache.

The startup June announced a $20 million pre-seed round, led by Marc Benioff's Time Ventures, with Michael Dell, Aaron Levie and George Kurtz also participating. Its platform scans a company's existing systems, maps where business processes stall, and then builds agent-driven replacements that connect into legacy software such as Salesforce, ServiceNow and Workday. Co-founder Efrat Rapoport argues that AI paradoxically increases demand for professional services, and pitches June as the step-by-step guide enterprises are missing.