Signum
Feed
Useful signal17 Nov 2026medium confidence

OpenAI releases GPT-6.2 Astra and Sol, cutting Astra API price 20% and claiming benchmark gains; independent index ranks Astra second behind Claude Fable 5.2

Edited by · written with AI from the linked sources (how)

OpenAI released two models, GPT-6.2 Astra (professional/science/law/finance work) and GPT-6.2 Sol (fast everyday tasks), to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and via the API. Astra API pricing is $8 input / $0.40 cached / $40 output per million tokens (20% below its predecessor); Sol pricing is unchanged at $2 / $0.10 / $10. Pro-200 subscribers gain Ultrafast mode access starting with Sol (Pro-500 already had it). Both models hit the 'Critical' cybersecurity threshold in OpenAI's Preparedness Framework, with stricter safeguards added and access for vetted researchers planned via OpenAI Daybreak. Artificial Analysis ranks Astra 62 (second to Claude Fable 5.2 at 65) and Sol 57, but reports flat or worse coding-agent results (Sol down 3 points) and a high hallucination rate.

CapabilityEconomicsAccessAdoption

Entities: OpenAI, GPT-6.2 Astra, GPT-6.2 Sol, Anthropic, Claude Fable 5.2, Artificial Analysis

67Useful signal
3 sources
0 primary
Was this useful?
01

What happened

OpenAI released two models, GPT-6.2 Astra (aimed at professional, science, law and finance work) and GPT-6.2 Sol (fast everyday tasks). They are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and through the API. Astra's API price is now $8 input / $40 output per million tokens, 20% below its predecessor. Sol's price is unchanged at $2 / $10. Pro-200 subscribers also get Ultrafast mode, starting with Sol. Artificial Analysis ranks Astra at 62 (second to Claude Fable 5.2 at 65) and Sol at 57, but reports flat or worse coding-agent results (Sol down 3 points) and a high hallucination rate.

02

Why it matters

Teams that use OpenAI models for document-heavy work in law, finance or research get a cheaper option, and Astra's second place in an independent index makes it a credible one to test against Claude Fable 5.2. Coding-agent users have little reason to switch, since the independent results are flat or worse. The high hallucination rate matters most for the professional uses Astra is pitched at, because errors there are costly and a lower price does not offset them. Both models were rated 'Critical' for cybersecurity under OpenAI's own framework. That means tighter safeguards, and access for vetted researchers only, so some security teams may find them restricted.

03

What is noise

'World's best model for scientists' and 'dominating the cost-performance frontier' are OpenAI's own claims, and most of the benchmark evidence comes from OpenAI too. The independent index puts Astra behind Anthropic's model, which weakens the 'on par' framing. A 20% price cut on one model is a normal point-release adjustment, not a market shift. Missing context is how the cost comparison holds up once hallucination and retry costs are included.

04

Watch next

  1. 01Independent evaluations on your own tasks, especially hallucination and error rates on legal, financial and scientific work, as opposed to OpenAI's chosen benchmarks.
  2. 02Whether Anthropic or Google (Gemini 4 Carbon) cut prices or ship updates in the next few weeks, and whether Artificial Analysis moves Astra's score or ranking after more testing.
  3. 03Details of the 'Critical' cybersecurity safeguards and the OpenAI Daybreak access programme: who qualifies, how long approval takes, and whether legitimate security work gets blocked.

Coverage

3 stories

Newsletter

Get the no-hype AI news straight to your inbox

Weekly. No spam, unsubscribe any time. Privacy

More capability signals

Full feed →