OpenAI puts out GPT-6.2 Astra and Sol, Nvidia won't take Huang back
OpenAI's GPT-6.2 models, Nvidia's board on Jensen Huang, AI picking military targets, a Java bug Codex found, Microsoft's coding model, Alibaba's Qwen4-Max and a report on the US outages.
OpenAI put out GPT-6.2 Astra and Sol, with Astra 20% cheaper Astra is the first model OpenAI has released that crosses its “High” risk level for helping to improve AI itself, and Sol is the cheaper one for everyday work. Both went live on Tuesday in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users, and on the API, where Astra costs $8 per million tokens in and $40 out. Source
Nvidia’s board says it won’t bring Jensen Huang back Chair Harvey Jones said on Tuesday’s earnings call that the board “does not intend to reinstate Jensen as CEO”, after Huang had posted earlier that day that he “would be honored” to return. Interim chief executive Colette Kress said on the same call that Nvidia will no longer sell its H200 chips in China. Source
Claude picked an Iranian school to strike in 69% of tests In tests Vals, an AI testing company, published on Tuesday, Claude Opus 5.5 chose the school in Minab, where a US strike in February killed more than 150 civilians, more often than any other model. Three OpenAI models never chose it until told it was a game. Anthropic told TechCrunch this use is “very clearly against our rules” and that it bans users it catches. Source
Codex found Java4Shell, a critical Java bug, for under $200 It can let an attacker take over a server running recent versions of Java on certain Intel and AMD chips, with apps like Jira, Jenkins and Keycloak affected. Quang Luong at the security firm Calif found it in about an hour using OpenAI’s Codex. Oracle shipped fixed versions on Tuesday, the day Calif made it public. Source
Microsoft put its own coding model, MAI Code 2, in Copilot Microsoft says it gets about 96 per cent of Claude Opus 5.5’s results across four coding tests, mostly run by Microsoft itself, at about 5 per cent of the cost. A lighter Flash version for quick, routine jobs came with it, and both went into GitHub Copilot on Tuesday at Microsoft’s Ignite conference. Source
Alibaba’s Qwen4-Max has 2.4 trillion parameters, and it isn’t open Only about 71 billion of them are working at any one time. It went on sale through Alibaba’s QwenCloud on Monday, and Alibaba says smaller Qwen4 models that anyone can download are coming soon. Source
AI-found bugs are behind the US outages, the Washington Sun says Three unnamed sources told it that bank executives and the NSA have privately told members of Congress the recent bank, Microsoft and NSA outages came from fixing flaws found by AI. The Sun says none looks like an attack. In public, the NSA has blamed scheduled maintenance, Chase a software bug and Microsoft a technical defect. The report came out on Tuesday. Source
Newsletter
Get The Daily Debrief by email
The last 24 hours in AI, every weekday.
You're on the list.
One last step: tap the link we've just emailed you. It's how we make sure nobody can sign you up but you.