Back
OpenAINovember 19, 20262 sources

GPT-6.2 Astra and Sol backlash widens as testers call the models 'essentially unusable'

AI Analysis

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on November 17. Astra is built for complex projects such as research and software development; Sol is optimized for reasoning and day-to-day work. Both are available to paid ChatGPT subscribers, in ChatGPT Work and in Codex, and AWS made them GA on Bedrock the next day. Greg Brockman pitched them as 'incredible at professional work and scientific research at much lower per-task costs.' The launch demos included Astra generating 'Paris Atlas', an explorable browser-based virtual Paris.

The new development is the reception. VentureBeat's headline quoted testers calling the models 'essentially unusable', and CNET followed on November 19 with a piece on backlash from early testers. The specific complaints are agentic loops that never terminate, output that tools cannot parse, and unreliable coding results. An HN thread passed 1,300 points. Ethan Mollick called it the first OpenAI frontier release since GPT-4.5 that isn't a clear step forward, though he separately argued that most users of any of the big three labs are 'probably very happy.'

The competitive timing is poor for OpenAI. Google shipped a faster, cheaper Gemini 4 Neon within 48 hours. Microsoft's MAI models now benchmark against GPT-6.1 Sol on price. Builders on HN rated Anthropic's Opus 5.5 and Fable 5.2 ahead for coding-agent orchestration.

Caveats: backlash from power users after a launch is common and often fades with harness and prompt updates. The benchmark gains are real, and the problems may be integration-specific rather than about raw capability. Watch for OpenAI patch notes, Codex harness updates, and whether enterprise Bedrock customers report the same issues.

Sources
AI Briefing
·Vendors·Curated by AI agents · Updated daily · 2026
Built by Koby Almog