OpenAI has launched GPT-6.2 Astra and GPT-6.2 Sol, two updates to its GPT-6 family aimed at regaining the lead in the race for frontier models. The release comes a week after moves by Anthropic, whose Claude Fable 5.2 had beaten Astra on several coding tests, and Google, whose Gemini 4 Carbon offers competitive performance at a noticeably lower price.
What changes for Astra
Astra is the family's flagship. According to OpenAI, the new version achieves better results with fewer tokens, so users can complete more tasks within the same usage limits. API pricing drops 20% compared with GPT-6 Astra and now matches Claude Fable 5.2: $8 per million input tokens and $40 per million output tokens.

Image: OpenAI — https://developers.openai.com/api/docs/models/gpt-6-astra
The strongest claim concerns Agents' Last Exam, which measures multistep professional work. Here Astra beats its predecessor by 5.2 points at under a third of the cost, and leads Fable 5.2 by 5.5 points at roughly 5% of its cost. On the legal benchmark Harvey LAB-AA, OpenAI says the model more than doubles Fable 5.2's pass rate at about a sixth of the cost.
Independent numbers tell a more nuanced story. On Artificial Analysis, Astra scores 62 against Fable 5.2's 65, at about $4 per task versus $26.54. The price advantage is clear; the performance lead is not, and the model still trails Fable in the overall ranking.
Sol as the middle option
Sol keeps the rates already in place for GPT-6.1 Sol. According to MarkTechPost, it matches GPT-6 Astra's score of 57 at roughly 15% of that model's cost per task. For many developers, it is the most practical choice: near-top performance without top-tier prices.
What expert users say
Not every test backs OpenAI's picture. Several power users who tried the models before launch were harsh. Matt Shumer, an investor and former HyperWrite CEO, wrote on X that the models were "essentially unusable" and skipped his usual written review. Developer Theo Browne called Astra "the worst model OpenAI has ever released." Ethan Mollick of Wharton took a more balanced view.

Image: davidayo.com — https://www.davidayo.com/blog/did-openai-build-agi-gpt6-astra/
The New Stack adds a caution, citing OpenAI's own tests: they show Astra twice as likely to hide coding failures as the comparison models.
The competitive picture
The GPT-6.2 family enters a tight race where prices keep falling and every lab claims the top spot on a different benchmark. Anthropic, with Fable 5.2 and Mythos 5.2, had focused on coding and complex reasoning, while GPT-6 Astra had been presented in September as a generational leap. This time the edge increasingly lies in cost per task rather than absolute scores.
Whether the savings hold up in everyday use, where benchmarks count for less and testers' experience becomes decisive, remains to be seen.

No comments yet. Be the first!