OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Tuesday, and within hours an independent test ranked Astra behind only Anthropic's Claude Fable 5.2, while finishing the same work for about 85% less per task even though the two charge identical token prices.
That gap is the story. Astra and Fable 5.2 both list at $8 per million input tokens and $40 per million output tokens. But Fable 5.2 writes far more to get to an answer, and you pay for every word.
The short answer: GPT-6.2 Astra scored 62 on the Artificial Analysis Intelligence Index, three points behind Fable 5.2, at $4.00 per task against $26.54. GPT-6.2 Sol scored 57 at $1.01 per task. The catch is coding, where neither model improved on the independent lab's agent tests.
OpenAI announced the pair on X a little before 1pm ET on Tuesday.
Some of you have been asking very politely. Some of you have just been asking.
— OpenAI (@OpenAI) November 17, 2026
GPT-6.2 Astra and Sol are now available.
Super efficient and built to perform where it counts—the real world.
https://t.co/qr2iP0GpfX pic.twitter.com/jLCyVIws2Y
What are GPT-6.2 Astra and GPT-6.2 Sol?
They are the two new models in OpenAI's lineup. Astra is the frontier model, pitched at "your hardest projects," and Sol is the cheaper, faster everyday model, according to OpenAI's launch post. Both replace earlier versions: Astra follows GPT-6 Astra and Sol follows GPT-6.1 Sol.
OpenAI's main pitch is efficiency. It says both models get better results "with fewer output tokens" (the chunks of text a model generates, which is what you're billed for), so the same paid plan or API budget stretches further.
The company leans hard on office work. It says Astra's slides, spreadsheets and documents "read less like AI" and follow a company's templates more closely. Early customers quoted in the post say the efficiency is real. "Our average time for long-horizon accounting work fell by 50%," said Eros Wang, a member of technical staff at Basis.
How much does GPT-6.2 Astra cost?
OpenAI cut Astra's price and kept Sol's flat, per its launch post:
- GPT-6.2 Astra: $8 per million input tokens, $0.40 per million cached input tokens (text the model has already seen in a conversation) and $40 per million output tokens. That's 20% below GPT-6 Astra on input and output, with cached input down 60%.
- GPT-6.2 Sol: $2 input, $0.10 cached input and $10 output per million tokens, the same as GPT-6.1 Sol.
In ChatGPT, both models are rolling out to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex starting Tuesday. Developers can call them in the API as gpt-6.2-astra and gpt-6.2-sol. OpenAI's faster Ultrafast mode is available on both for Pro 500 subscribers, and Pro 200 subscribers get it for the first time, starting with Sol.
How does GPT-6.2 Astra compare with Claude Fable 5.2, Gemini 4 Carbon and Opus 5.5?
Artificial Analysis, an independent lab that runs every major model through the same set of tests, published its results on launch day. Here is where the new models land on its Intelligence Index, with the lab's measured cost and time to finish each task, as of Wednesday evening:
- Claude Fable 5.2: 65, $26.54 per task, 41.2 minutes.
- GPT-6.2 Astra: 62, $4.00 per task, 13.7 minutes.
- Gemini 4 Carbon: 60, $4.90 per task, 36.1 minutes.
- Claude Opus 5.5: 59, $10.04 per task, 24.6 minutes.
- GPT-6.2 Sol: 57, $1.01 per task, 8.5 minutes.
- GPT-6 Astra: 57, $6.24 per task, 20.0 minutes.
- GPT-6.1 Sol: 55, $1.27 per task, 23.4 minutes.
The cost gap with Fable 5.2 comes down to verbosity. Fable 5.2 generated 275 million output tokens to get through the index, according to Artificial Analysis. GPT-6.2 Astra needed 65.8 million, roughly a quarter as much. Same price per token, a quarter of the tokens, and you land at about 15% of the bill.
Speed follows the same pattern. Astra finished the average task in about a third of the time Fable 5.2 took, and Sol was the fastest model in the group.
Sol may be the better deal for most people. It matches GPT-6 Astra's score at about 16% of its cost per task, and it's about 20% cheaper per task than GPT-6.1 Sol while scoring two points higher, per its Artificial Analysis page. Artificial Analysis said OpenAI's models now occupy most of its cost-per-task Pareto frontier (the set of models where you can't get a higher score without paying more), shared only with Anthropic and Meta.
You can put any of these side by side in our AI model comparison tool, which has pages for GPT-6.2 Astra, GPT-6.2 Sol and Claude Fable 5.2.
Where are GPT-6.2 Astra and Sol weaker?
Coding. On the Artificial Analysis Coding Agent Index, measured in OpenAI's Codex harness at maximum effort, Astra was flat with GPT-6 Astra and Sol lost three points against GPT-6.1 Sol, the lab said. It traced most of Sol's drop to a 14-point fall on one question-answering test, where the new model was more likely to leave out key facts.
That lines up with something OpenAI flagged itself. In its safety notes, the company said both models were more likely than earlier ones to skip mentioning obstacles that blocked their work, such as a missing file, in two tests. It said it added instructions in Codex that "substantially reduce" those omissions and is still investigating.
Artificial Analysis also measured hallucination, meaning how often a model guesses instead of admitting it doesn't know a factual answer. Astra scored 59% and Sol 55% at maximum effort, so you still need to check their facts.
OpenAI also said Astra is its first publicly released model to reach the "High" level for AI self-improvement under its Preparedness Framework, which OpenAI likens to giving every one of its researchers an extra, highly capable research engineer. Both models meet its "Critical" threshold for cybersecurity, as GPT-6 Astra did.
What does OpenAI claim GPT-6.2 Astra does best?
On OpenAI's own charts, the biggest wins are in long professional tasks and science. The company says Astra beat GPT-6 Astra by 5.2 points on Agents' Last Exam (a test of multistep professional work) at less than a third of the cost, and topped Fable 5.2 there by 5.5 points. It also says Astra set a new high on Terminal-Bench-Science, a test of scientific work done in a computer terminal, and called it "the best model in the world for scientists."
Artificial Analysis partly backs that up: it found Astra leading its own Terminal-Bench-Science results and its 3D-building test, and second only to Fable 5.2 on a hard physics reasoning set. The other benchmark figures in OpenAI's post are the company's own.
Should you switch to GPT-6.2 Astra or Sol?
If you're on GPT-6.1 Sol through the API, switching to GPT-6.2 Sol looks like a free upgrade on most work: same token price, higher score, lower cost per task. Test your coding agents before you move them, since that's the one place the independent numbers went backward.
If you're paying for Claude Fable 5.2, Astra is now the obvious thing to test against it. Fable 5.2 still has the top score, but Astra gets within three points for about 15% of the cost per task and a third of the wait. Google's Gemini 4 Carbon now sits below Astra on the index. For lighter jobs, our breakdown of GPT-6.1 Terra pricing covers OpenAI's cheaper tier, and our AI API cost calculator will show what your own token volumes cost before you move a workload.
GPT-6.2 Astra FAQ
When was GPT-6.2 Astra released?
OpenAI released GPT-6.2 Astra and GPT-6.2 Sol on Tuesday, Nov. 17, 2026, in ChatGPT and the OpenAI API.
How much does GPT-6.2 Astra cost in the API?
$8 per million input tokens, $0.40 per million cached input tokens and $40 per million output tokens. GPT-6.2 Sol costs $2, $0.10 and $10.
Is GPT-6.2 Astra better than Claude Fable 5.2?
Not on raw score. Artificial Analysis gave Fable 5.2 a 65 and Astra a 62. Astra costs about 85% less per task in that test and finishes tasks about three times faster.
Who can use GPT-6.2 Astra in ChatGPT?
Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex. Enterprise and Edu workspaces need an admin to turn it on.
OpenAI didn't take the top spot on Tuesday. It did something that may matter more to anyone paying the bill: it made near-top performance cheap.


