Updated
Updated · OpenAI · Nov 17
OpenAI Launches GPT-6.2 Astra and Sol, Cutting Astra API Input Prices 20%
Updated
Updated · OpenAI · Nov 17

OpenAI Launches GPT-6.2 Astra and Sol, Cutting Astra API Input Prices 20%

3 articles · Updated · OpenAI · Nov 17

Summary

  • GPT-6.2 Astra and GPT-6.2 Sol are available starting today in ChatGPT Work, Codex and the API, with OpenAI positioning Astra for complex professional work and Sol for everyday tasks.
  • 5.2 points on Agents’ Last Exam put Astra ahead of GPT-6 Astra at less than one-third the cost, while Sol came within 0.5 points of Astra on FrontierCode 2.0 at about one-fifth the cost.
  • 20% lower Astra input and output token prices, plus a 60% cached-input discount, set API rates at $8 input, $0.40 cached input and $40 output per million tokens; Sol keeps prior pricing at $2, $0.10 and $10.
  • 9 percentage points on Terminal-Bench-Science, 6.2 points on GDPval v2 and a 1,318 Elo lead on AA-3D underpin OpenAI’s pitch that Astra better handles science, business documents, coding and interactive design.
  • High preparedness thresholds for biological, chemical and AI self-improvement risks, plus Critical cyber capability, led OpenAI to ship both models with tighter safeguards that may limit some legitimate security research.

Insights

Will the cheaper GPT-6.2 Sol quietly become OpenAI's most dominant weapon in crushing rival models for everyday enterprise tasks?
What hidden AI self-improvement risks did Astra exhibit to trigger unprecedented external safety audits and potential launch delays?
Could OpenAI's push for prompt caching actually be a desperate attempt to mask the skyrocketing compute costs of frontier models?