OpenAI's GPT-6.2 Astra Lists at $8 per Million Input Tokens, 20% Under Its Predecessor, While Coding Scores Stay Flat
A.I. / news
OpenAI's GPT-6.2 Astra Lists at $8 per Million Input Tokens, 20% Under Its Predecessor, While Coding Scores Stay Flat
Artificial Analysis ranks Astra second at 62 behind Claude Fable 5.2, and Sol's coding-agent score fell 3 points.

OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Nov. 17, pricing Astra at $8 per million input tokens and $40 per million output tokens, 20% below GPT-6 Astra. Sol keeps GPT-6.1 Sol's price of $2 and $10, according to The Decoder's account of the launch.
OpenAI's launch post returned an HTTP 403 error to this reporter's fetch on Nov. 18. The figures below come from OpenAI's own API model pages, which were readable, and from The Decoder, which reported the launch and the outside evaluations that followed it.
GPT-6.2 Astra and Sol pricing and limits
Both models list a 1,050,000-token context window, 922,000 tokens of maximum input and 128,000 tokens of maximum output, per OpenAI's Astra and Sol pages. Astra's knowledge cutoff is May 18, 2026. Sol's is June 1, 2026.
| Model | Input per 1M | Cached input per 1M | Output per 1M |
|---|---|---|---|
| GPT-6.2 Astra | $8 | $0.40 | $40 |
| GPT-6.2 Sol | $2 | $0.10 | $10 |
Requests above 272,000 input tokens double the input and cache rates and raise output to 1.5 times the base rate for the whole request. Batch jobs cost half. Fast mode costs twice the standard rate. Ultrafast costs three times through Feb. 28, 2027, then six times from March 1, 2027.

What the independent evaluations found
Artificial Analysis scored Astra 62 on its Intelligence Index, second to Claude Fable 5.2 at 65, according to The Decoder. Astra is up 5 points on GPT-6 Astra. Sol scored 57, up 2 points.
- Claude Fable 5.265 points
- GPT-6.2 Astra62 points
- Gemini 4 Carbon60 points
- Claude Opus 5.559 points
- GPT-6.2 Sol57 points
Source: Artificial Analysis, as reported by The Decoder, accessed 2026-11-18
Coding did not move. On Artificial Analysis's Coding Agent Index, run in the Codex harness at maximum compute, Astra matches GPT-6 Astra and Sol drops 3 points from GPT-6.1 Sol. The Decoder attributes the drop to a 14-point fall on SWE-Atlas-QnA-AA, where Sol omits key facts more often. On the AA-Omniscience hallucination test at maximum effort, Astra's rate is 59% and Sol's is 55%. Artificial Analysis notes the questions are unusually hard.
Epoch AI gave Astra an Epoch Capabilities Index score of 174, tying Fable 5.2. On real-world research tasks Epoch said Astra trails Fable 5.2 and barely improves on GPT-6 Astra. Astra solved 7 of 68 FrontierMath Erdős tasks, up from the record of 3 set by GPT-6.1 Sol.
Vals AI ranked Astra 7th and Sol 9th on its Vals Index. Astra costs $10.68 per task against $21.28 for GPT-6 Astra. Sol costs $3.18 per task, the cheapest model in Vals's top 10.
Which numbers are OpenAI's own
The benchmark claims OpenAI made itself are vendor-supplied and were not independently reproduced here. They include Astra's 5.2-point gain on Agents' Last Exam over GPT-6 Astra at under a third of the cost, and a 6.2-point gain on GDPval-v2, which The Decoder labels internal. OpenAI also reports that Astra clears a 50% success rate on ExploitBench with safeguards off, another internal test.
Legal results split by evaluator. OpenAI says Astra more than doubles Fable 5.2's pass rate on Harvey LAB-AA. Vals ranked Astra 11th on its own HLAB legal benchmark and 30th on Legal Research.
Safety designations and availability
OpenAI said both models reach the Critical cybersecurity threshold in its Preparedness Framework, as GPT-6 Astra did. It said Astra is the first externally released model to reach the High threshold for AI self-improvement. OpenAI plans to route security-research access through Daybreak.
The models are in ChatGPT Plus, Pro, Enterprise and Edu, in Codex and in the API. Pro-200 gets Ultrafast mode for the first time, starting with Sol.
OpenAI has not published weights for either model. The next data point is Artificial Analysis's cost-per-task figures for Astra against Fable 5.2, which The Decoder's account did not include.
For earlier open-weight comparisons on the same index, see Qwen3.8-27B's slow, verbose rating, and for how much the model choice drives agent cost, Runta's 27,280-run harness evaluation.
Sources
More in A.I.
- 01Gemini 4 Carbon Takes 36 Minutes a Task; Kernel Counts 7x the Steps of AstraArtificial Analysis puts Carbon at 36.0 minutes per Intelligence Index task against a 12.4-minute median, and a Kernel study traces most of the gap to extra steps.
- 02Researcher Says Claude Fable 5.2 Learned Protein Sequences; Test Shows a 45-Point GapXinming Tu's probe of 668 obscure bacterial proteins finds the newest Claude models predict real sequences far better than shuffled ones. GPT, Gemini and open-weight models do not.
- 03Tencent's Hy4-flash Weights Carry a License That Requires a Separate Deal Above $50 Million in RevenueThe 12B-active model is ungated on Hugging Face, but it ships under the Tencent Hy License, not Apache 2.0.
- 04Google Put Gemini 4 Hydrogen Behind Search for Everyone on Nov. 12, and Published No Click FiguresThe one randomised study of AI Overviews found a 39.8% drop in outbound clicks on early-2026 Search.