OpenAI released GPT-6.2 Sol and GPT-6.2 Astra on 17 November 2026. GPT-6.2 Sol keeps GPT-6.1 Sol’s API price of $2 per million input tokens and $10 per million output tokens, while GPT-6.2 Astra is priced 20% below GPT-6 Astra, at $8 and $40.
In ChatGPT, both models are in Work and Codex, not in regular Chat. OpenAI’s announcement names Plus, Pro, Enterprise and Edu plans. One OpenAI pricing page also lists Business, so Business access is unclear as of 18 November 2026.
This post is for L&D operations leads and AI tools admins who already run workflows on GPT-6.1 Sol or GPT-6 Astra and need to decide whether to switch. It is based on OpenAI’s published documentation. This site did not test the models, so every performance figure below is vendor-reported.
What did OpenAI release on 17 November 2026?
OpenAI released two models on 17 November 2026: GPT-6.2 Astra (API ID gpt-6.2-astra), pitched at the hardest projects, and GPT-6.2 Sol (gpt-6.2-sol), pitched at everyday coding and agent work. Both arrived in the API, ChatGPT Work and Codex that day, according to OpenAI’s announcement thread on its developer forum.
Both models share a 1,050,000-token context window, a 128,000-token maximum output, text and image input and text output, per OpenAI’s API models page. Knowledge cutoffs are June 1, 2026 for Sol and May 18, 2026 for Astra.
Two things are not confirmed. OpenAI’s own announcement page could not be opened for this post, so the forum copy is the source for plan and benchmark claims. And no page that could be read gives a retirement date for GPT-6.1 Sol or GPT-6 Astra. OpenAI’s API changelog for 17 November lists no deprecations.
How much does GPT-6.2 Sol cost in the API?
GPT-6.2 Sol costs $2.00 per million input tokens, $0.10 for cached input and $10.00 for output, the same as GPT-6.1 Sol. GPT-6.2 Astra costs $8.00, $0.40 and $40.00. These are Standard-tier rates on OpenAI’s API pricing page as of 18 November 2026.
Several modifiers apply to both new models:
- Long context: over 272,000 input tokens, input and cache rates double and output costs 1.5 times the short-context rate for the whole request.
- Batch: 50% of Standard, so $1 and $5 for Sol and $4 and $20 for Astra.
- Fast mode: 2 times Standard.
- Ultrafast: 3 times Standard through February 28, 2027, then 6 times from March 1, 2027.
- Background windows (public beta): 30% below Standard for a two-hour window and 40% below for an eight-hour window.
The pricing page lists background windows for the GPT-6.2 models, GPT-6.1 Terra and GPT-6 Luna. It does not list them for GPT-6.1 Sol or GPT-6 Astra. For overnight jobs such as bulk quiz-item drafting, that is a real difference.
Per-token price is not per-task price, so here is a worked example. Assumption: a job uses 20,000 input tokens and 4,000 output tokens, with no caching, and each model uses the same number of tokens. GPT-6 Astra costs $0.40 for that job and GPT-6.2 Astra costs $0.32. Sol costs $0.08 on either version.
OpenAI says GPT-6.2 Astra costs under one-third as much per task as GPT-6 Astra. A 20% list-price cut cannot produce that alone, so OpenAI must be counting fewer tokens per task (an inference, not an OpenAI statement). Your own prompts decide whether you see it.
Measure Cost Per Accepted Output
Run the same 20 prompts on the old and new model, log tokens for each run, then divide total spend by the number of outputs your reviewers accept without edits. A cheaper model that needs two review passes can cost more than a pricier one.
How does GPT-6.2 Sol differ from GPT-6.1 Sol?
On published specifications, GPT-6.2 Sol differs from GPT-6.1 Sol in knowledge cutoff, Ultrafast terms, background-window pricing and estimated Codex usage. Price, context window and maximum output are unchanged. The table uses only figures from OpenAI’s pages, read on 18 November 2026.
| Item | GPT-6.1 Sol | GPT-6.2 Sol |
|---|---|---|
| API model ID | gpt-6.1-sol | gpt-6.2-sol |
| Input / cached / output per 1M tokens | $2.00 / $0.10 / $10.00 | $2.00 / $0.10 / $10.00 |
| Context window / max output | 1,050,000 / 128,000 | 1,050,000 / 128,000 |
| Knowledge cutoff | April 30, 2026 | June 1, 2026 |
| Ultrafast API rate | 6 times Standard | 3 times Standard through Feb 28, 2027, then 6 times |
| Ultrafast included usage in Work and Codex | 8 times | 6 times |
| Estimated Codex messages per 5 hours (Plus, Standard Business) | 15 to 160 | 17 to 180 |
| Background windows (beta) | Not listed on pricing page | 30% or 40% below Standard |
| Reasoning effort | low, medium (default), high, xhigh, max | Levels not listed on model page; none is rejected |
No head-to-head benchmark numbers for the two Sol versions could be found in the readable text. OpenAI’s announcement says only that Sol nearly matches Astra on FrontierCode 2.0 at about one-fifth the cost. Treat “better than 6.1 Sol” as untested for your workload.
How does GPT-6.2 Astra differ from GPT-6 Astra?
GPT-6.2 Astra is cheaper than GPT-6 Astra on every Standard-tier price line OpenAI publishes, by 20% on input and output and 60% on cached input. It also has a later knowledge cutoff and a different Ultrafast arrangement. Context window and maximum output are unchanged.
| Item | GPT-6 Astra | GPT-6.2 Astra |
|---|---|---|
| API model ID | gpt-6-astra | gpt-6.2-astra |
| Input / cached / output per 1M tokens | $10.00 / $1.00 / $50.00 | $8.00 / $0.40 / $40.00 |
| Cache write (30 minutes) | $12.50 | $10.00 |
| Batch input / output | $5 / $25 | $4 / $20 |
| Fast input / output | $20 / $100 | $16 / $80 |
| Context window / max output | 1,050,000 / 128,000 | 1,050,000 / 128,000 |
| Knowledge cutoff | April 30, 2026 | May 18, 2026 |
| Ultrafast included usage in Work and Codex | 8 times | 6 times |
| Estimated Codex messages per 5 hours (Plus, Standard Business) | 5 to 45 | 6 to 50 |
The Codex message figures come from OpenAI’s Work and Codex pricing page, which calls them estimates, not fixed limits. They are small upgrades, not step changes.
Which ChatGPT plans get GPT-6.2 Sol and Astra?
Plus, Pro, Enterprise and Edu users get both models in ChatGPT Work and Codex, per OpenAI’s announcement. Business is listed on OpenAI’s Work and Codex pricing page but not in the announcement. Neither model was added to regular Chat. Free and Go are not listed for either model.
| Plan | GPT-6.2 Sol and Astra in Work and Codex | Notes |
|---|---|---|
| Free and Go | Not listed | The pricing page shows only GPT-6 Luna for these plans. |
| Plus | Yes, per announcement | No Ultrafast. |
| Pro | Yes, per announcement | Ultrafast on Pro $500 for both models. OpenAI says it is coming to Pro $200, starting with Sol. |
| Business | Conflicting: pricing page lists it, announcement does not | One Business Premium user reported not seeing the models; the reply called this normal during rollout. Not confirmed. |
| Enterprise and Edu | Yes, per announcement | Off by default until a workspace admin enables them, per the pricing page. |
A reply in the announcement thread, which appears to come from OpenAI staff, says availability depends on plan, workspace permissions and rollout progress, and that users should update the desktop app and Codex CLI. For how the plans compare on price, see ChatGPT pricing by plan. For how Sol, Luna and Astra split work, see GPT-6 Sol vs Luna vs Astra.
What changes for API integrations moving to GPT-6.2?
Four API behaviors change or need checking when you move to GPT-6.2, according to OpenAI’s latest-model guide. None changes the price, but each can change output quality or cost if your code assumes the old behavior.
- Edited history drops reasoning: the Responses API drops reasoning items after mid-conversation edits to history. The guide says quality can fall and token use can rise, and advises keeping the prefix unchanged.
- Instructions move: top-level instructions are appended after input. For reusable, cacheable instructions, the guide recommends a developer message as the first input item.
- Effort settings: requests that set effort to none are rejected. The guide suggests starting new integrations at medium and moving none or minimal users to low.
- Style and parameters: the guide says Astra is more concise and direct by default, so revisit “be concise” instructions. It also lists temperature and top_p as parameters to omit.
Tool calling requires the Responses API, per OpenAI’s changelog. If your L&D tooling still calls Chat Completions with tools, plan that change first.
What has OpenAI claimed about GPT-6.2 benchmarks?
OpenAI has published few readable numbers. All of the following are vendor-reported, and no independent reproduction was found. Astra at xhigh effort scores 5.2 points above GPT-6 Astra at max effort on Agents’ Last Exam, and jumps 22 points on ExploitBench Internal Port. Sol comes close to Astra on FrontierCode 2.0 at about one-fifth the cost.
The announcement’s other benchmark results sit in charts that could not be read as text, so they are not quoted here. None of these tests measures instructional design, quiz quality or compliance accuracy.
OpenAI also says it launched both models with tighter safeguards for higher-risk cybersecurity assistance. Assumption: this could change how the models answer security-awareness or incident-response training prompts. Re-test those prompts first.
Should you switch from GPT-6.1 Sol or GPT-6 Astra?
Sol users can test freely because the price is unchanged. Astra users have a clearer reason to test, since list prices fell 20%. No retirement date turned up for either older model, so there is no deadline. Switch a workflow only after it passes your own checks.
Use this checklist, in order, for each workflow:
- Check plan access. Confirm the model appears in your plan and, for Enterprise and Edu, that an admin has enabled it. Do not assume Business access.
- Freeze a prompt set. Pick 20 to 30 real prompts and save their current outputs from the older model as the baseline.
- Re-run and compare. Score outputs against the same rubric. Re-check anything that relied on “be concise” or on sampling parameters.
- Audit API behavior. Look for edited conversation history, top-level instructions, none effort settings and Chat Completions tool calls.
- Recalculate cost per task. Use your logged tokens, not list prices, and include review time.
- Check usage headroom. Compare the estimated message ranges for your plan against your team’s real volume.
- Plan a rollback. OpenAI lists no dated snapshots for either 6.2 model, so you cannot pin one version. Keep the older model ID in your configuration while it stays available.
Test Assessed Content First
Sort your prompts by consequence before you re-test. Anything feeding graded assessments, compliance modules or SCORM and xAPI-tracked courses gets human review of every output. Drafts, summaries and internal job aids can pass on a smaller sample.
For ideas on which learning jobs suit the larger model, see GPT-6 Astra for L&D.
What should you do next?
Pick one low-risk workflow that runs on GPT-6.1 Sol or GPT-6 Astra today and run the checklist on it this week. Keep the old model live while you compare, and write down the cost per accepted output for both.
Check OpenAI’s pricing page and your workspace settings again before you commit. Rollout is still in progress, Business access is unresolved and Ultrafast pricing changes on March 1, 2027.
FAQ
Q1. When was GPT-6.2 Sol released?
OpenAI released GPT-6.2 Sol, together with GPT-6.2 Astra, on 17 November 2026, according to its announcement thread and API changelog. The models arrived in ChatGPT Work and Codex and in the API on the same day, with the API model IDs gpt-6.2-sol and gpt-6.2-astra.
Q2. How much does GPT-6.2 Sol cost per million tokens?
As of 18 November 2026, OpenAI lists GPT-6.2 Sol at $2.00 per million input tokens, $0.10 for cached input and $10.00 for output on the Standard tier. That matches GPT-6.1 Sol. Requests over 272,000 input tokens double the input rate and raise output to 1.5 times the normal rate.
Q3. Is GPT-6.2 Sol available in regular ChatGPT Chat or on the free plan?
Not as of 18 November 2026. A reply in OpenAI’s announcement thread says the models were not added to regular Chat. They are in ChatGPT Work and Codex for Plus, Pro, Enterprise and Edu users. Free and Go plans are not listed for either model, and Business access is not confirmed.
Q4. Is GPT-6.2 Astra cheaper than GPT-6 Astra?
Yes, on list price. GPT-6.2 Astra costs $8 per million input tokens and $40 for output, compared with $10 and $50 for GPT-6 Astra. Cached input falls from $1.00 to $0.40. OpenAI also claims under one-third the cost per task, a vendor figure you should check against your own prompts.
Q5. Do I need to change my code to move from GPT-6.1 Sol to GPT-6.2 Sol?
Possibly. OpenAI’s guide says the Responses API drops reasoning items after mid-conversation edits to history, appends top-level instructions after the input, and rejects a reasoning effort of none. Tool calling requires the Responses API. If your code does none of these things, the switch may be only a model ID change, but test first.
Q6. Will OpenAI retire GPT-6.1 Sol and GPT-6 Astra?
No retirement date was found. OpenAI’s API changelog for 17 November lists no deprecations, and the model pages for GPT-6.1 Sol and GPT-6 Astra show no deprecation notice. One forum user reported seeing GPT-6.1 labeled legacy in a picker, which is an unverified user report. Recheck OpenAI’s deprecations page before you plan around either model.