Key takeaways
- GPT-6.2 Sol replaces GPT-6.1 Sol at the same $2 input and $10 output per million tokens, with a June 1, 2026 knowledge cutoff instead of April 30.
- There is no GPT-6.1 Astra: GPT-6.2 Astra replaces GPT-6 Astra and cuts the price from $10 and $50 to $8 and $40.
- Artificial Analysis scores GPT-6.2 Astra 62, behind only Claude Fable 5.2, and Epoch AI has the two effectively tied at 174.17 and 173.5.
GPT-6.2 vs 6.1 comes down to this: GPT-6.2 Sol replaces GPT-6.1 Sol at the same price, $2 per million input tokens and $10 per million output tokens, with a newer knowledge cutoff, cheaper Ultrafast mode and better scores. There was never a GPT-6.1 Astra. The flagship skipped the 6.1 step, so GPT-6.2 Astra replaces GPT-6 Astra and costs 20% less: $8 and $40 instead of $10 and $50. OpenAI released both 6.2 models on November 17, 2026, according to its API changelog.
The context window did not change: every model here reads up to 1,050,000 tokens and writes up to 128,000.
This comparison is based on OpenAI's model pages, its system card and published leaderboard results, not on hands-on testing. For the launch itself, see our news article on GPT-6.2 Astra and Sol. For choosing between the two new models, see GPT-6.2 Astra vs Sol.
Which 6.1 models GPT-6.2 replaces
OpenAI names its models by tier: Astra is the top model, Sol the everyday one, Terra and Luna smaller. Each tier updates on its own schedule:
- GPT-6 Astra (
gpt-6-astra), September 3, 2026. - GPT-6 Sol and GPT-6 Luna, September 22.
- GPT-6.1 Sol (
gpt-6.1-sol), September 29. - GPT-6.1 Terra (
gpt-6.1-terra), October 26. - GPT-6.2 Astra and GPT-6.2 Sol, November 17.
So "GPT-6.1" means two models: Sol and Terra. GPT-6.2 Sol is the direct upgrade to GPT-6.1 Sol. GPT-6.2 Astra is the upgrade to GPT-6 Astra, because a gpt-6.1-astra model page does not exist (it returns a 404). There is no 6.2 Terra, so GPT-6.1 Terra ($1 in, $5 out) stays the cheap option.
GPT-6.2 vs 6.1: prices and limits side by side
Prices are per million tokens, for prompts up to 272,000 input tokens, from OpenAI's model pages for GPT-6.2 Sol, GPT-6.1 Sol, GPT-6.2 Astra and GPT-6 Astra, as of November 18, 2026.
| GPT-6.2 Sol | GPT-6.1 Sol | GPT-6.2 Astra | GPT-6 Astra | |
|---|---|---|---|---|
| Released | Nov 17, 2026 | Sep 29, 2026 | Nov 17, 2026 | Sep 3, 2026 |
| Input | $2.00 | $2.00 | $8.00 | $10.00 |
| Cached input | $0.10 | $0.10 | $0.40 | $1.00 |
| Cache write, 30 minutes | $2.50 | $2.50 | $10.00 | $12.50 |
| Output | $10.00 | $10.00 | $40.00 | $50.00 |
| Ultrafast mode | 3x standard until Feb 28, 2027, then 6x | 6x standard | 3x standard until Feb 28, 2027, then 6x | Price not listed |
| Context window | 1,050,000 | 1,050,000 | 1,050,000 | 1,050,000 |
| Max input / output | 922,000 / 128,000 | 922,000 / 128,000 | 922,000 / 128,000 | 922,000 / 128,000 |
| Knowledge cutoff | June 1, 2026 | April 30, 2026 | May 18, 2026 | April 30, 2026 |
| Input and output | Text and images in, text out | Same | Same | Same |
| Artificial Analysis index (max effort) | 57 | 55 | 62 | 57 |
| Epoch Capabilities Index | Not listed yet | 168.63 | 174.17 | 169.2 |
The same rules apply to all four for the extras. Requests over 272,000 input tokens pay double for input and cache and 1.5 times for output, on the whole request. Batch jobs cost half. Fast mode costs twice the standard rate. All four support the Responses API, Chat Completions and Batch, the same built-in tools (web search, file search, code interpreter, computer use, MCP, hosted shell and others), structured outputs and prompt caching. None of them supports fine-tuning, audio or video.
What changed from GPT-6.1 Sol to GPT-6.2 Sol
- Newer knowledge. The cutoff moves from April 30 to June 1, 2026.
- Cheaper Ultrafast. Ultrafast mode, OpenAI's fastest serving option, cost six times the standard rate on GPT-6.1 Sol. On GPT-6.2 Sol it costs three times until February 28, 2027, then six times from March 1.
- A 12-hour cache. The 6.2 Sol page lists a 12-hour cache write at $5 per million tokens, next to the 30-minute one at $2.50. OpenAI added 12-hour caching on October 20.
- More work per plan. OpenAI's ChatGPT changelog says that with GPT-6.2 Sol "you can complete more tasks on the same paid plan than with GPT-6.1 Sol."
One behavior change applies to both 6.2 models. OpenAI's GPT-6.2 guide says that starting with GPT-6.2, the API drops the model's saved reasoning when you edit earlier turns of a conversation, switch to a smaller model or change the user or organization. Losing it can cost quality and tokens, so keep the start of a conversation unchanged when you change settings.
What changed from GPT-6 Astra to GPT-6.2 Astra
Astra got cheaper on every line. Input and output fall 20%, from $10 and $50 to $8 and $40. Cached input falls 60%, from $1.00 to $0.40, which matters most for agents that resend the same long context many times. The knowledge cutoff moves from April 30 to May 18, 2026.
The guide also says Astra now writes more concisely by default, so prompts that tell it to "be concise" may need another look.
Benchmarks OpenAI published
OpenAI's launch charts plot score against cost without printing most exact numbers, and its announcement page could not be read for this article. Exact numbers are in the GPT-6.2 system card, published the same day. It is a safety document, so its tests lean toward cyber, biology and AI research. It is also the only OpenAI source that puts GPT-6.2 Sol next to GPT-6.1 Sol:
| Test (OpenAI's results) | GPT-6.2 Sol | GPT-6.1 Sol | GPT-6.2 Astra | GPT-6 Astra |
|---|---|---|---|---|
| SEC-Bench Pro, reproducing security bugs | 88.34% | 78.8% | 86.75% | 85.4% |
| ExploitBench Internal Port, turning browser bugs into working exploits | 50.5% | 21.5% | 53.5% | 31.5% |
| GPQA Diamond, graduate science questions, no tools | 96.0% | Not reported | 97.0% | 96.0% |
| Internal Research Debugging | 74.66% | Not reported | 81.85% | 78.05% |
| KernelGen 1P, writing GPU kernels | 74.71% | Not reported | 76.92% | 66.72% |
| MLE-Bench Revised, mean percentile | 91.38 | Not reported | 96.06 | 93.80 |
| PostTrainBench Lite, share of gap closed | 44.89% | Not reported | 74.79% | 76.97% |
The biggest jump from 6.1 to 6.2 is on exploit writing: Sol more than doubled its score. That is why OpenAI rates both 6.2 models "Critical" for cybersecurity, the top of its scale, and launched them with tighter limits on higher-risk security requests. Astra is rated "High" for AI self-improvement, the first model OpenAI has released at that level.
Astra does not win everywhere: GPT-6 Astra still leads on PostTrainBench Lite. All of these are OpenAI's own tests, several of them internal, so treat them as the company's claims.
How independent leaderboards rank GPT-6.2 Astra
As of November 18, 2026, two independent groups put GPT-6.2 Astra near the top, but not clearly first.
Artificial Analysis. On its Intelligence Index v5.0.1, which combines nine evaluations, Claude Fable 5.2 at max effort leads with 65. GPT-6.2 Astra at max effort scores 62, tied with Fable 5.2 at xhigh effort. No other model scores higher, so Astra trails only Fable 5.2. The site's GPT-6.2 Astra page ranks it third of 92 entries, where Fable 5.2 counts once for each effort setting.
The gains over the older models are clearest here:
| Artificial Analysis, max effort | Index score | Cost per index task | Output speed |
|---|---|---|---|
| GPT-6.2 Astra | 62 | $4.00 | 64 tokens/s |
| GPT-6 Astra | 57 | $6.24 | 44 tokens/s |
| GPT-6.2 Sol | 57 | $1.01 | 85 tokens/s |
| GPT-6.1 Sol | 55 | $1.27 | 49 tokens/s |
GPT-6.2 Sol matches GPT-6 Astra's score at about a sixth of the cost per task. GPT-6.2 Astra scores 5 points more than GPT-6 Astra and costs about a third less per task. Artificial Analysis now marks GPT-6.1 Sol and GPT-6 Astra as deprecated; OpenAI's own model pages carry no deprecation notice.
Epoch AI. On the Epoch Capabilities Index, which combines many benchmarks run the same way on every model, GPT-6.2 Astra scores 174.17 and Claude Fable 5.2 scores 173.5. Their ranges overlap almost entirely (170.29 to 178.58 for Astra, 170.09 to 177.36 for Fable 5.2), so the two are effectively tied. Gemini 4 Carbon follows at 169.78, then GPT-6 Astra at 169.2, Claude Opus 5.5 at 168.7 and GPT-6.1 Sol at 168.63. GPT-6.2 Sol was not on Epoch's list yet.
Availability in ChatGPT and the API
In the API, all four models are live. GPT-6.2 Astra and Sol are gpt-6.2-astra and gpt-6.2-sol, and the older model pages are still up with no shutdown date.
In ChatGPT, OpenAI's launch post says Plus, Pro, Enterprise and Edu users get both 6.2 models in ChatGPT Work and Codex, its coding agent. OpenAI's ChatGPT changelog names only ChatGPT Work and Codex, not the regular chat. Pro subscribers at $200 a month can use Ultrafast with GPT-6.2 Sol, and Pro subscribers at $500 a month can use it with both models. The launch post does not list the Business plan, and it does not say when the 6.1 models leave the model picker.
Which to use
For anyone on GPT-6.1 Sol, switching is close to free: same price, newer knowledge, a higher Artificial Analysis score and a lower cost per task there. The main work is testing prompts that depend on how the model writes, and keeping conversation prefixes stable so reasoning isn't dropped.
For anyone on GPT-6 Astra, GPT-6.2 Astra costs less and scores higher on both independent indexes. Workloads that resend cached context save the most.
Frequently asked questions
Is there a GPT-6.1 Astra?
No. The 6.1 generation has only GPT-6.1 Sol and GPT-6.1 Terra. GPT-6.2 Astra is the successor to GPT-6 Astra, released September 3, 2026.
Is GPT-6.2 Sol more expensive than GPT-6.1 Sol?
No. Both cost $2 per million input tokens, $0.10 for cached input and $10 per million output tokens. Ultrafast mode is cheaper on GPT-6.2 Sol, at three times the standard rate until February 28, 2027, against six times on GPT-6.1 Sol.
Is GPT-6.2 better than Claude Fable 5.2?
Not clearly. Artificial Analysis scores Fable 5.2 at 65 and GPT-6.2 Astra at 62 on its Intelligence Index, while Epoch AI's Capabilities Index has Astra at 174.17 and Fable 5.2 at 173.5, within each other's margin of error.
Can I still use GPT-6.1 Sol?
Yes, in the API. As of November 18, 2026, OpenAI's GPT-6.1 Sol model page shows no deprecation notice or shutdown date.
Sources
- API changelogOpenAI
- GPT-6.2 Sol model pageOpenAI
- GPT-6.1 Sol model pageOpenAI
- GPT-6.2 Astra model pageOpenAI
- GPT-6 Astra model pageOpenAI
- GPT-6.1 Terra model pageOpenAI
- Using GPT-6.2OpenAI
- GPT-6.2 Astra and GPT-6.2 Sol system cardOpenAI
- ChatGPT changelogOpenAI
- Artificial Analysis Intelligence IndexArtificial Analysis
- GPT-6.2 Astra: intelligence, performance and price analysisArtificial Analysis
- GPT-6.2 Sol: intelligence, performance and price analysisArtificial Analysis
- GPT-6.1 Sol: intelligence, performance and price analysisArtificial Analysis
- GPT-6 Astra: intelligence, performance and price analysisArtificial Analysis
- Epoch Capabilities IndexEpoch AI
