Quick Answer
GPT-6.2 Astra and Sol are OpenAI’s newest models, launched November 17, 2026. Astra is the frontier model for professional work (science, law, finance, research); Sol is the cheaper everyday model. Astra costs $8 per million input tokens and $40 per million output tokens (cached input $0.40), Sol costs $2 and $10 (cached input $0.10). Both have a 1,050,000-token context window and model ids gpt-6.2-astra and gpt-6.2-sol. They are available in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu subscribers, and in the API. They are not added to regular ChatGPT Chat.
What Are GPT-6.2 Astra and Sol?
GPT-6.2 is a two-tier release, not a single model. Astra is pitched as OpenAI’s strongest model for professional work — financial models, spreadsheets, slide decks, research and scientific work. Sol is the efficient sibling for everyday tasks, at a quarter of Astra’s price on input and output.
The split matters because the two models are close on some benchmarks and far apart on price. On Cognition’s FrontierCode 2.0, both pass 48.2% of tasks, but Sol costs $1.29 per run against Astra’s $5.97 (Modelflare’s figures). On abstract reasoning (ARC-AGI-3), Astra leads by 15.4 percentage points. So the choice is not “better vs worse” in general — it is which kind of work, and how much each run is worth.
Both models take text and image input, with a 1,050,000-token context window, up to 922,000 input tokens and 128,000 output tokens per call, and no fine-tuning. Astra’s knowledge cutoff is May 18, 2026; Sol’s is June 1, 2026.
Release Date and Availability
OpenAI released GPT-6.2 Astra and Sol on November 17, 2026, publishing its safety system card the same day. The launch came about seven weeks after GPT-6.1 Astra was cancelled before release.
They are available now through:
- ChatGPT Work and Codex — Plus, Pro, Enterprise and Edu subscribers
- API — model ids
gpt-6.2-astraandgpt-6.2-sol, in the Responses API and Chat Completions (tool calling needs the Responses API) - Amazon Bedrock — generally available
- GitHub Copilot — Pro+, Max, Business and Enterprise plans, billed at provider list prices
They are not in regular ChatGPT Chat: an OpenAI staff reply on the developer forum says the rollout does not add either model there, and availability depends on plan, workspace permissions and rollout progress.
Key GPT-6.2 Features
1. Two tiers for two kinds of work
Astra for demanding professional and scientific work; Sol for everyday tasks. The tiering is the point of the release — pay frontier prices only when the work needs them.
2. A 1.05-million-token context window
Both models share the same 1,050,000-token window, so long documents, large codebases and multi-file research fit in a single call.
3. Cheaper cached input on Astra
Astra’s cached input fell 60% to $0.40 per million tokens (from $1). Sol’s cached input is $0.10. Repeated system prompts and long shared documents get much cheaper on the second call.
4. Discounted background windows
In the Responses API public beta, a two-hour completion window costs 30% below Standard and an eight-hour window 40% below, on Astra, Sol, GPT-6.1 Terra and GPT-6 Luna.
5. Ultrafast output mode, widened
OpenAI is expanding Ultrafast, a faster output mode: $500/month Pro subscribers get it on both models, and $200 Pro subscribers get it for the first time, starting with Sol.
6. A usage-limit change coming in December
From December, Plus, Pro and Business plans will share one allowance across ChatGPT and Codex, measured in tokens — so larger models and longer tasks use more of it. Astra and Sol are pitched as getting more done before subscribers reach that allowance.
GPT-6.2 Pricing Explained
Standard rates, per million tokens, for short-context requests (up to 272,000 input tokens). Above 272,000 input tokens, input and cache rates double and output costs 1.5x the short-context rate.
| Item | Astra | Sol |
|---|---|---|
| Input tokens | $8.00 | $2.00 |
| Output tokens | $40.00 | $10.00 |
| Cached input tokens | $0.40 | $0.10 |
| 30-min cache write | $10.00 | $2.50 |
| Change vs previous | 20% below GPT-6 Astra | Unchanged from GPT-6.1 Sol |
A worked example: a task using 1 million input tokens and 500,000 output tokens costs $8 + (0.5 × $40) = $28 on Astra, and $2 + (0.5 × $10) = $7 on Sol — a quarter of Astra’s price. With cached input, the same Astra task’s input portion falls to $0.40.
One price coincidence worth knowing: Astra now lists at exactly the same $8/$40 as Anthropic’s Claude Fable 5.2, which Anthropic cut by 20% a week earlier. List price is not cost per finished task, though — independent testers found Astra’s cost per task fell 36% (Artificial Analysis) and 50% (Vals AI) versus GPT-6 Astra, well short of OpenAI’s own “under one-third” claim (see Benchmarks).
GPT-6.2 Astra vs Sol
| Spec | Astra | Sol |
|---|---|---|
| API model id | gpt-6.2-astra |
gpt-6.2-sol |
| Best for | Science, law, finance, research | Everyday tasks |
| Context window | 1,050,000 tokens | 1,050,000 tokens |
| Max input / output | 922,000 / 128,000 tokens | 922,000 / 128,000 tokens |
| Input price per 1M | $8.00 | $2.00 |
| Output price per 1M | $40.00 | $10.00 |
| Cached input per 1M | $0.40 | $0.10 |
| Knowledge cutoff | May 18, 2026 | June 1, 2026 |
| FrontierCode 2.0 | 48.2% pass, $5.97/run | 48.2% pass, $1.29/run |
Performance and Benchmarks
All performance figures here are the vendors’ own unless marked independent, and independent testers disagree with OpenAI on the size of the saving.
OpenAI’s own claims: Astra beats GPT-6 Astra by 5.2 points on Agents’ Last Exam at under a third of the cost, and leads Claude Fable 5.2 by 5.5 points at 5% of the cost per task. On its GDPval v2 test (multistep consulting, finance and design assignments), Astra scores 6.2 points higher than GPT-6 Astra while using 22% fewer output tokens.
Independent measurements are more modest. Artificial Analysis puts Astra at $4.00 per task on its Intelligence Index at maximum effort, against $6.24 for GPT-6 Astra — a 36% drop. Vals AI measures $10.68 per task against $21.28 — about a 50% drop. Artificial Analysis ranks Astra second on its Intelligence Index at 62, behind Claude Fable 5.2 at 65; Sol scores 57. The figures come from different test sets, so none contradicts another outright — but a team budgeting from OpenAI’s number would expect a bigger cut than either outside tester found.
Early tester reactions reported by VentureBeat were mixed to negative: Matt Shumer called the models essentially unusable, and Ethan Mollick found Astra uneven and harder to steer. Treat that as early reaction, not a settled verdict.
Safety Ratings
OpenAI’s system card, published on launch day, rates both models at the Critical level for cybersecurity and at the High level for biological and chemical risks under its Preparedness Framework. Astra meets the High threshold for AI self-improvement; Sol falls below it. OpenAI tightened default limits on higher-risk cyber help and requires extra vetting for looser access.
How to Use GPT-6.2 Astra & Sol
1. Pick the tier by the work, not the name
Use Astra for science, law, finance, research and long-horizon professional work. Use Sol for everyday tasks — especially coding tasks where both models pass at the same rate, since Sol costs about a fifth as much per run.
2. Use ChatGPT Work or Codex if you are on a paid plan
Plus, Pro, Enterprise and Edu subscribers get both models in ChatGPT Work and Codex. Do not look for them in regular ChatGPT Chat — they are not there.
3. Call the model by its id in the API
Use gpt-6.2-astra or gpt-6.2-sol in the Responses API or Chat Completions. Tool calling needs the Responses API.
4. Cache repeated prompts
Put instructions and long documents you reuse into cached input — $0.40 per million tokens on Astra instead of $8, $0.10 instead of $2 on Sol.
5. Use background windows for slow jobs
If a job can wait two or eight hours, the Responses API background windows cut the price by 30% or 40%.
Best Uses
- Professional work (Astra) — financial models, spreadsheets, slide decks, research, scientific work
- Everyday work (Sol) — routine writing, coding and everyday tasks at a quarter of Astra’s price
- Long documents and codebases — the 1.05M-token window fits both in a single call
- Agents — tool calling through the Responses API, with background windows for slow jobs
Limitations and Things to Know
- Not in regular ChatGPT Chat. The models are in ChatGPT Work and Codex only, for paid plans.
- No fine-tuning. Neither model supports it at launch.
- The “under one-third” cost claim is OpenAI’s own. Independent testers measured 36–50% cost-per-task drops, not the two-thirds-plus cut OpenAI claims.
- Safety ratings are the highest tier for cyber. Both models are rated Critical for cybersecurity under OpenAI’s Preparedness Framework, with tighter default limits on higher-risk cyber help.
- Long-context rates jump. Above 272,000 input tokens, input and cache rates double and output costs 1.5x — check the threshold before sending very long prompts.
- Usage limits change in December. Plus, Pro and Business plans will share one token-measured allowance across ChatGPT and Codex.
Frequently Asked Questions
What is GPT-6.2 Astra & Sol?
GPT-6.2 Astra and Sol are OpenAI’s newest models, launched November 17, 2026. Astra is the frontier model for demanding professional work; Sol is the cheaper everyday model. Both share a 1.05-million-token context window.
How much does GPT-6.2 cost?
Astra costs $8 per million input tokens and $40 per million output tokens (cached input $0.40). Sol costs $2 and $10 (cached input $0.10). Both are 20% / unchanged versus their predecessors — Astra is 20% below GPT-6 Astra, Sol keeps GPT-6.1 Sol’s price.
Where can I use GPT-6.2 Astra & Sol?
In ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu subscribers, through the API (gpt-6.2-astra / gpt-6.2-sol), on Amazon Bedrock, and in GitHub Copilot for Pro+, Max, Business and Enterprise. Not in regular ChatGPT Chat.
What is the difference between Astra and Sol?
Astra is for demanding professional and scientific work; Sol is for everyday tasks at a quarter of Astra’s price. On FrontierCode 2.0 both pass 48.2% of tasks, but Sol costs $1.29 per run against Astra’s $5.97. On abstract reasoning (ARC-AGI-3) Astra leads by 15.4 points.
What is the context window?
1,050,000 tokens for both models, with up to 922,000 input tokens and 128,000 output tokens per call.
Is GPT-6.2 cheaper than GPT-6?
Astra’s list price is 20% lower than GPT-6 Astra. Independent testers measured cost per finished task falling 36% (Artificial Analysis) to 50% (Vals AI) — less than OpenAI’s own “under one-third” claim.
What are the model ids?
gpt-6.2-astra and gpt-6.2-sol.
Is GPT-6.2 safe?
OpenAI’s system card rates both models Critical for cybersecurity and High for biological and chemical risks, with tighter default limits on higher-risk cyber help. Astra also meets the High threshold for AI self-improvement.
Final Verdict
GPT-6.2 is a pricing and tiering release more than a capability jump: Astra gets 20% cheaper per token, Sol stays cheap, and the real decision is which tier your work needs. The honest picture is split — OpenAI’s own benchmarks claim large gains at a fraction of the cost, while independent testers found a real but smaller 36–50% cost-per-task drop, and early reactions were mixed. If your work is everyday coding and writing, Sol at $2/$10 is the value pick; if it is science, law or finance, Astra is the frontier option — but budget from the independent numbers, not the headline claim, and remember neither model is in regular ChatGPT Chat.
Editorial update: November 19, 2026. This article is based on OpenAI’s launch announcement and system card (17 Nov 2026), OpenAI’s API pricing page, and secondary coverage. Performance figures are OpenAI’s own unless marked independent, and have not been independently verified by faqsnow.