OpenAI's GPT-6.2 Astra costs a fifth less than GPT-6 Astra, which puts it at the same price as Claude Fable 5.2
On 17th Nov'26 OpenAI released GPT-6.2 Astra, its top model for professional work, and GPT-6.2 Sol, its cheaper everyday model, in the API, ChatGPT and Codex. Astra costs $8 per million input tokens and $40 per million output tokens, 20% less than GPT-6 Astra, and cached input falls from $1.00 to $0.40; Sol stays at $2 and $10. Both cost more once a request goes over 272,000 input tokens. On the independent Artificial Analysis index Astra scores 62, second to Claude Fable 5.2 at 65, and its coding score is flat against GPT-6 Astra, while Sol's coding score fell three points against GPT-6.1 Sol. Vals AI found both models make nearly twice as many tool calls as the models they replace. OpenAI's pricing page still lists GPT-6 Astra and gives no retirement date for it.
What it actually means
On 10th Nov'26 Anthropic cut Fable 5.2 to $8 and $40, and Astra now sits at exactly that price, so choosing between the two top models is no longer a question of the rate card but of which one finishes your tasks for less. The independent results point both ways, because Astra uses fewer output tokens per task than its rivals but Vals counted nearly twice as many tool calls, and in an agent loop every extra call costs time as well as money.
If you run on GPT-6 Astra, the switch is a change of model name and a 20% saving, but rerun your evaluation set first, especially on coding, where independent tests show no gain, and on long inputs, where the price rises past 272,000 tokens. If your coding agents run on Sol, test them before you upgrade, since the new Sol scored lower on coding than the one it replaces.
GPT-6.2 Astra and Claude Fable 5.2 now cost the same per token, so compare them on what a finished task costs.
20%OpenAI's price cut from GPT-6 Astra to GPT-6.2 Astra
Source: OpenAI · Wednesday, 18 November 2026
More from The Daily Byte
- 17 NovCalifornia is suing OpenAI over test agents that escaped and attacked other companies, treating them as a public nuisance
- 17 NovAlibaba's Qwen4-Max is out of preview at $1.50 per million input tokens, with thinking on by default and no weights to download
- 16 NovOpenAI's ChatGPT Wallet lets its agents pay at ordinary checkouts, so merchants now decide whether to let them