GPT-6.1 Sol is OpenAI’s new middle model, and it nearly matches GPT-6 Astra on agentic coding, computer use and professional work at one-fifth of Astra’s token prices. It replaced GPT-6 Sol just seven days after that model launched. In the video below I test it inside Codex and ChatGPT. Here’s what it costs, how it scores, how to access it and when you should still reach for Astra.
Key takeaways
- GPT-6.1 Sol costs $2 in / $10 out per million tokens, vs $10 / $50 for Astra.
- 75.2% on DeepSWE v1.1 at high effort, ahead of Astra’s 74.1% at xhigh.
- 1.05M context window and up to 128K output tokens.
- Available in ChatGPT Work, Codex and the API. Claim a free strategy session to plan your setup.
What is GPT-6.1 Sol?
OpenAI now has three main GPT-6 models. Astra is the smartest and most expensive. Luna is the fastest and cheapest. Sol sits in the middle. OpenAI’s launch post says GPT-6.1 Sol “nearly matches GPT-6 Astra’s intelligence on agentic coding, computer use, and professional work” at one-fifth of Astra’s prices.
GPT-6 Astra arrived on 3 September 2026. GPT-6 Sol and Luna followed on 22 September. GPT-6.1 Sol was published on 29 September, so the original Sol lasted about a week.
| Model | Input | Output | Cached input |
|---|---|---|---|
| GPT-6 Astra | $10.00 | $50.00 | $1.00 |
| GPT-6.1 Sol | $2.00 | $10.00 | $0.10 |
| GPT-6 Luna | $0.10 | $0.50 | $0.01 |
Cached input at $0.10 is 95% cheaper than normal input, and half the cached price of the old GPT-6 Sol.
GPT-6.1 Sol benchmarks
OpenAI published cost per task next to each score, which is the useful bit. Here are the main results:
| Benchmark | GPT-6.1 Sol | Compared with |
|---|---|---|
| DeepSWE v1.1 (coding) | 75.2% high ($0.65/task) | Astra xhigh 74.1% ($4.43) |
| GDP.pdf (professional work) | 32.0% high ($0.35) | Astra xhigh 32.2% ($1.91), Opus 5.5 high 28.8% |
| AutomationBench 1.0.6 | 31.7% medium ($0.19) | Opus 5.5 29.5%, GPT-6 Sol 26.9% |
| OSWorld 2.0 (computer use) | 71.4% max ($1.27) | Astra max 73.5% ($9.44) |
| Terminal-Bench Science 0.1 | 57.0% max ($5.47) | Astra 68.1% ($23.80), Opus 5.5 63.3% |
The pattern is clear. On coding and office work, GPT-6.1 Sol matches or nearly matches Astra for a fraction of the cost per task. On computer use it’s within 2.1 points of Astra at about a seventh of the cost.
Science is the exception. On Terminal-Bench Science, Astra is still well ahead. OpenAI itself points people to Astra for the hardest science work.
Want GPT-6.1 Sol working in your business? Inside the AI Profit Boardroom you get the Agent OS — Hermes, OpenClaw, OpenCode, Claude and Codex in one place — plus step-by-step Hermes training and four live coaching calls a week.
Want a plan for your business first? Claim your free strategy session.
Join the AI Profit Boardroom →Claim Your Free SEO Strategy Session →
Fewer factual mistakes
OpenAI also tested factual errors on deliberately hard prompts. At low effort, the share of answers with a factual mistake fell from 11.4% to 7.7%, a drop of about a third. Across settings, it stays within 1.9 points of Astra at less than a fifth of the cost per task. OpenAI also published a system card addendum with its safety results.
Want to know which GPT-6 model fits your business? Claim your free strategy session →
Who can use GPT-6.1 Sol?
- ChatGPT Work and Codex on Plus, Pro, Business, Enterprise and Edu plans.
- Enterprise and Edu: an admin has to switch it on.
- Free and Go plans weren’t included at launch.
- The API as
gpt-6.1-sol. - Ultrafast mode in Codex, with up to 8× faster token generation, was announced for “the coming days” and is listed for Pro and eligible Enterprise and Edu plans.
It isn’t in the standard ChatGPT chat yet — it’s in ChatGPT Work and Codex. If you mainly want it for coding, I wrote a separate guide: GPT-6.1 Sol for coding.
Want the short version? I also wrote my GPT-6.1 Sol overview on the AI Profit Boardroom blog.
GPT-6.1 Sol API specs
From the model page:
| Spec | GPT-6.1 Sol |
|---|---|
| Context window | 1,050,000 tokens |
| Max input | 922,000 tokens |
| Max output | 128,000 tokens |
| Knowledge cutoff | 30 April 2026 |
| Input / output | Text and image in, text out |
| Endpoints | Responses, Chat Completions (no tool calling), Batch |
| Batch and Flex | 50% cheaper |
| Fast / Ultrafast | 2× / 6× the standard price |
| Prompts over 272K | 2× input price, 1.5× output price |
Built-in tools include web search, file search, image generation, code interpreter, hosted shell, apply_patch, skills, computer use, MCP and tool search. For tool calling, use the Responses API.
Effort levels and migration tips
GPT-6.1 Sol has five reasoning efforts: low, medium (the default), high, xhigh and max. Unlike GPT-6 Sol, there’s no “none” or “minimal”. OpenAI’s migration guide says to move any none calls to low, and to remove temperature, top_p and top_logprobs from your requests.
- Use low for quick, simple jobs — it scored 64.4% on DeepSWE at $0.17 per task.
- Use medium for most work.
- Use high or above for hard coding and research.
- Save max for jobs where getting it right matters more than cost.
What I found in my test
In the video I tried it in Codex and ChatGPT. It handled one-prompt 3D worlds well enough, but they were less polished than what I got from Sonnet 5.5. Where it shines is value. Astra can burn through your usage fast, while GPT-6.1 Sol gets most of the way there for much less.
GPT-6.1 Sol vs Astra vs Luna: which one?
| If you need… | Pick |
|---|---|
| The hardest science and research problems | GPT-6 Astra |
| Coding, computer use and office work at a good price | GPT-6.1 Sol |
| Huge volumes of simple tasks for the lowest cost | GPT-6 Luna |
For most businesses, GPT-6.1 Sol is the sensible default. It’s five times cheaper than Astra and close on most tests. Luna is twenty times cheaper again, so use it for simple, high-volume jobs like sorting, tagging and short replies. Only pay for Astra when Sol clearly can’t do the job, and test that on your own tasks rather than trusting any benchmark table, including mine.
A quick way to decide: run the same five real tasks from your business through Sol and Astra. If Sol gets four or five right, you’ve just cut that part of your bill by about 80%. If it misses the hard ones, send only those to Astra and keep everything else on Sol.
My verdict on GPT-6.1 Sol
GPT-6.1 Sol is the GPT-6 model most people should use day to day. It nearly matches Astra on coding, computer use and office work, it’s a fifth of the price, and it has a huge 1M-token context. Keep Astra for the hardest science problems, and use GPT-6.1 Sol for everything else.
About Julian Goldie
I’m Julian Goldie — founder of Goldie Agency and the AI Profit Boardroom. I test new AI agents almost every day for 432K+ YouTube subscribers and run Hermes, OpenClaw, OpenCode, Claude and Codex side by side in my own Agent OS. Claim your free strategy session or join the AI Profit Boardroom.
FAQ: gpt-6.1 sol
What is GPT-6.1 Sol?
OpenAI’s middle GPT-6 model, sitting between Astra and Luna. It nearly matches Astra on coding, computer use and professional work at a fifth of the price.
How much does GPT-6.1 Sol cost?
$2 per million input tokens, $10 per million output tokens and $0.10 for cached input.
Is GPT-6.1 Sol better than GPT-6 Astra?
Not overall. It matches Astra on DeepSWE coding and nearly matches it on office work and computer use, but Astra is still clearly better at hard science tasks.
Can I use GPT-6.1 Sol in free ChatGPT?
Not at launch. It’s in ChatGPT Work and Codex on Plus, Pro, Business, Enterprise and Edu plans, and in the API.
What is the GPT-6.1 Sol context window?
1,050,000 tokens, with up to 922,000 input tokens and 128,000 output tokens.
Ready to put Hermes to work in your business? Join the AI Profit Boardroom for the Agent OS, setup training and daily updates.
Or claim your free strategy session and we’ll map out your setup.
Starting from zero? Get the free AI course and community.
Join the AI Profit Boardroom →Claim Your Free SEO Strategy Session →Get the FREE AI Course + Community →
Related reading
Last updated October 2026. This page is a living guide to gpt-6.1 sol — the facts here move fast and it is updated as they do.
