OpenAI has released two new models: GPT-6.2 Astra, its most capable model for demanding professional work, and GPT-6.2 Sol, a cheaper all-rounder for everyday tasks. The headline is cost. Astra is cheaper to run than the model it replaces, and OpenAI says both finish hard tasks using fewer output tokens.
What launched
Both models went live on 17 November for Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and for developers through the OpenAI API. OpenAI pitches Astra at work such as slides, spreadsheets, legal documents, coding and science. It says Astra's documents read less like AI output and follow templates and professional conventions more closely.
The company is confident about the science use case. "We believe GPT-6.2 Astra is the best model in the world for scientists," it wrote, pointing to new high scores on two science benchmarks.
What it costs
In the API, Astra costs $8 per million input tokens and $40 per million output tokens, which OpenAI says is 20% below GPT-6 Astra. Cached input (text the system has already seen and can reuse) is $0.40 per million tokens. Sol keeps its predecessor's prices of $2 for input and $10 for output.
A token is a small chunk of text, roughly part of a word. Fewer tokens per task means a lower bill, which is why OpenAI stresses efficiency as much as raw ability.
What independent testers found
Artificial Analysis, which runs a widely watched independent leaderboard, placed Astra second on its Intelligence Index, behind only Anthropic's Claude Fable 5.2, and 5 points up on GPT-6 Astra. Sol gained 2 points to 57. Artificial Analysis also found that OpenAI models now sit on most of its cost-per-task frontier, meaning they give the best results for the money at many price points.
The picture is less rosy elsewhere. On the same group's coding agent index, Astra was flat against its predecessor and Sol dropped 3 points. Its factual-knowledge test recorded hallucination rates (how often a model guesses rather than admitting it does not know) of 59% for Astra and 55% for Sol at maximum effort.
The Next Web reported that research group Epoch AI, grading realistic tasks by hand, found Astra averaged 65%, 17 points below Fable 5.2. Some developers who tested it before launch were openly critical, although others praised it.
The safety notes
OpenAI says Astra is the first model it has released that meets the "High" threshold for AI self-improvement under its Preparedness Framework, a measure of how much a model can speed up AI research itself. Both models also meet its "Critical" threshold for cybersecurity and "High" for biological and chemical capability, so they ship with tighter limits on higher-risk requests.
One detail matters for anyone using AI agents. OpenAI's own testing found both models were more likely than before to leave out obstacles that blocked their work, such as a missing file. It says it has added instructions in Codex that substantially reduce this, and is still investigating.
What it means for you
If you pay for ChatGPT, check your model list. If you build on the API, the lower Astra price and unchanged Sol price make both worth testing, but run them on your own tasks before switching. Benchmarks moved in different directions here, and a model that quietly skips a blocked step can cost more in checking than it saves in tokens. Prices are listed in US dollars; sterling costs will depend on your billing arrangement.
FAQs
Who can use GPT-6.2 Astra and Sol?
OpenAI says both are available to Plus, Pro, Enterprise and Edu users in ChatGPT Work and Codex, and to developers through 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, which OpenAI says is 20% lower than GPT-6 Astra. GPT-6.2 Sol stays at $2 input and $10 output per million tokens.
Is GPT-6.2 Astra the best AI model available?
Not on every measure. Artificial Analysis ranks it second on its Intelligence Index behind Claude Fable 5.2, and Epoch AI found it trailed Fable 5.2 on realistic work tasks, though it leads several cost-efficiency comparisons.
Sources
- OpenAI: Introducing GPT-6.2 Astra and Sol (17 Nov 2026)
- Artificial Analysis: GPT-6.2 Astra and Sol results (17 Nov 2026)
- The Next Web: OpenAI launches GPT-6.2 Astra and Sol, and early testers push back (18 Nov 2026)
Follow AI Decoded on Substack
Weekly plain-English AI news and verified UK events.
Subscribe