Timeline

OpenAI releases GPT-6.2 Astra and GPT-6.2 Sol

Astra costs $8 input and $40 output per million tokens, 20% below GPT-6 Astra; ARC Prize verified 100.0% on ARC-AGI-3 with a provider-adapter harness, 70.3% with the standard one.

  • Models & capabilities
  • Labs & people
  • Notable

OpenAI released GPT-6.2 Astra, which it called its frontier model “for your hardest projects”, and GPT-6.2 Sol, its efficient everyday model, on 17 November, as point updates to GPT-6 Astra and GPT-6.1 Sol. According to OpenAI’s launch post, both were available the same day to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex and through the API as gpt-6.2-astra and gpt-6.2-sol. Astra costs $8 per million input tokens, $0.40 cached and $40 output, 20% below GPT-6 Astra; Sol keeps GPT-6.1 Sol’s $2, $0.10 and $10.

The launch claims are OpenAI’s own. On Agents’ Last Exam it said Astra set a new high, 5.2 points above GPT-6 Astra at under a third of the cost and 5.5 points ahead of Claude Fable 5.2 at 5% of the cost. A footnote says the leaderboard’s Fable 5.2 results exclude nine tasks that used a fallback model. OpenAI also reported new highs on Terminal-Bench-Science 0.1 and the Vals Science Index, and said Sol came within 0.5 points of Astra on FrontierCode 2.0 at about a fifth of the cost. It described the cyber capabilities as “another significant jump” and launched with tighter safeguards that, it said, restrict some legitimate vulnerability research.

Independent results were mixed. ARC Prize listed verified scores for Astra on 17 November: 98.5% on ARC-AGI-1, 95.8% on ARC-AGI-2, and on ARC-AGI-3 70.3% with its standard harness and 100.0% with a provider-adapter harness, which preserves the model’s reasoning state between requests and compacts long conversations. ARC’s published data lists Fable 5.2 at about 81.2% on the standard harness and about 97.7% on the provider-adapter one, so Astra led on one and trailed on the other. The live Artificial Analysis leaderboard, read on 22 November, put Fable 5.2 at about 65 on its Intelligence Index and Astra at about 62. Epoch AI’s separate real-work evaluation of the two is covered in its own entry.

OpenAI said both models meet its Critical threshold for cybersecurity and its High threshold for biological and chemical capabilities, and that Astra is its first externally released model to meet High for AI self-improvement. The system card carries those findings. The release came eight days after Anthropic’s Fable 5.2 and Mythos 5.2, and OpenAI named Fable 5.2 in several of its comparisons.

In the commentary

What people were saying around this time — external links, from the record's commentary rail.