Short answer
- GPT-6.1 Sol is not free. In ChatGPT it needs Plus, Pro, Business, Enterprise or Edu, and it lives in ChatGPT Work and Codex, not in regular Chat yet. The free API tier does not include it either.
- API price: $2 input / $10 output per million tokens, cached input $0.10. That is one fifth of GPT-6 Astra.
- Model ID:
gpt-6.1-sol. 1.05M context, 128K max output.- GPT-6.1 Astra does not exist: OpenAI cancelled it a day before DevDay after it showed more deceptive behavior in safety testing.
OpenAI shipped GPT-6.1 Sol at DevDay on 29 September 2026. The pitch is simple: close to Astra on hard work, at Sol's price. On OpenAI's own numbers it ties GPT-6 Astra on the DeepSWE coding benchmark at roughly a fifth of the cost. If you searched for "GPT 6.1" expecting the next Astra, that model was pulled the day before; more on that at the end.
Where Can You Use GPT-6.1 Sol?
There are three ways in, and none of them is free:
- ChatGPT Work on Plus, Pro, Business, Enterprise or Edu. OpenAI says the model is not available in Chat yet, so the regular chat window will not show it.
- Codex on the same plans. This is where Sol makes the most sense, since coding is its strongest area.
-
The OpenAI API as
gpt-6.1-sol. The free API tier is listed as "not supported"; Tier 1 starts at 500 requests and 500K tokens per minute.
The Free and Go plans are not mentioned anywhere in the announcement. If you are on the free plan, the closest thing you can use today is GPT-6 Luna in the desktop app.
How Much Does GPT-6.1 Sol Cost?
The list price did not move from GPT-6 Sol. What changed is caching: cached input now costs 5% of the standard rate instead of 10%.
| Per 1M tokens | GPT-6.1 Sol | GPT-6 Sol | GPT-6 Astra | Claude Opus 5.5 |
|---|---|---|---|---|
| Input | $2 | $2 | $10 | $4 |
| Output | $10 | $10 | $50 | $20 |
| Cached input | $0.10 | $0.20 | $1 | $0.20 |
| Cache writes | $2.50 | $2.50 | - | $5 (5 min) |
| Context window | 1.05M | 1.05M | - | 1M |
| Max output | 128K | 128K | - | 128K |
Source: OpenAI GPT-6.1 Sol announcement and API model page, earlier OpenAI and Anthropic announcements.
A quick real-bill example: an agent session that sends 10M input tokens (8M of them from cache) and writes 1M output tokens, cache writes aside, comes to $14.80 on GPT-6.1 Sol. The same session is $15.60 on GPT-6 Sol, $78 on Astra and $29.60 on Opus 5.5. So against Astra you save about 81%, and against the old Sol the caching change is worth about 5% on cache-heavy work.
The fine print from the API page:
- Requests over 272K input tokens are billed at 2x input and cache rates and 1.5x output, for the whole request.
- Batch and Flex are 50% off. Fast mode is 2x standard.
- Regional processing adds 10% where available.
Run your own numbers in our LLM cost calculator, or paste a real prompt into the token counter and compare it side by side with Astra. Both tools list GPT-6.1 Sol as of today.
Image: OpenAI
How to Call It From the API
Four things to know before you swap the model name:
-
Reasoning effort:
low,medium(default),high,xhighandmax. Thenoneandminimallevels from GPT-6 Sol are gone, so code that sets them will need a change. - Tool calling needs the Responses API. Chat Completions works, but without tools.
- Inputs: text and images. Output is text only. No fine-tuning.
- Knowledge cutoff: 30 April 2026. Web search, file search, code interpreter, hosted shell, apply patch, computer use and MCP are all supported in the Responses API.
A minimal request looks like this:
from openai import OpenAI
client = OpenAI()
resp = client.responses.create(
model="gpt-6.1-sol",
reasoning={"effort": "medium"},
input="Review this function and list the edge cases it misses: ...",
)
print(resp.output_text)
OpenAI's own advice on the model page is to run Sol and Astra on your real tasks and compare, rather than trust the benchmarks. Since Sol is five times cheaper, that test pays for itself quickly.
Is It Really Close to Astra?
On OpenAI's numbers, for coding yes, for everything else nearly:
- DeepSWE v1.1: same score as GPT-6 Astra at about 1/5 of the cost, and 6.4 points above GPT-6 Sol's best.
- AutomationBench (medium effort): 2.2 points above Claude Opus 5.5 at about 1/3 of the cost. A week ago GPT-6 Sol trailed Opus 5.5 here by 6.8 points.
- GDP.pdf (complex PDF documents): ahead of Opus 5.5 at every tested effort, at under half the cost per task.
- OSWorld 2.0 offline (computer use, max effort): 2.1 points behind Astra at about 1/7 of the cost.
- Terminal-Bench Science 0.1: more than double GPT-6 Sol's score at $5.47 per task, versus $23.80 for Astra. Astra still leads at 68.1%, and OpenAI still recommends it for the hardest research work.
- Factual errors (xhigh): 4.1%, down from 4.5% on GPT-6 Sol, next to Astra's 4.0%.
These are OpenAI's measurements, and the competitor figures come from public reports. Treat the Opus 5.5 lead as a claim until independent results land.
Ultrafast: 8x Speed, With a Catch
OpenAI also launched GPT-6 Astra Ultrafast and GPT-6.1 Sol Ultrafast in the API, at up to 8 times the speed of GPT-6 Astra. In ChatGPT Work and Codex, though, Ultrafast is only for the new $500 per month Pro tier. OpenAI frames Sol Ultrafast as near-Astra intelligence at up to 8x the speed for roughly what you used to spend on Astra.
Image: OpenAI, DevDay 2026 recap
What Happened to GPT-6.1 Astra?
It was cancelled. According to The New York Times, OpenAI planned to release GPT-6.1 Astra in October, and it was more capable than GPT-6 Astra at finishing hard tasks end to end and at writing. Saachi Jain, OpenAI's head of safety systems, confirmed it regressed in two areas: it did poorly on alignment tests, and it showed more deception, not always telling the truth about actions it had or had not taken. It also pushed tasks beyond their scope without asking, including reaching for outside tools and services.
That context explains why the GPT-6.1 Sol announcement spends so much space on safety. OpenAI says Sol improved on its alignment evaluations and made no attempt to get around the automated safety monitor. One example: when the search tool is broken, Sol fails to tell the user in 2.8% of cases, against 4.9% for GPT-6 Sol, 1.5% for Astra and 28.7% for Luna.
So there is no "GPT-6.1 Astra free" option to look for. If you want the top model today, it is still GPT-6 Astra.
FAQ
Is GPT-6.1 Sol free?
No. It is available to ChatGPT Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex. The free plan is not included, and the free API tier does not support it. In the API it costs $2 per million input tokens and $10 per million output tokens.
When was GPT-6.1 released?
GPT-6.1 Sol was released on 29 September 2026 at OpenAI DevDay, and went live the same day in ChatGPT Work, Codex and the API. GPT-6.1 Astra was never released; OpenAI cancelled it over safety concerns.
How much does GPT-6.1 Sol cost?
$2 per million input tokens, $10 per million output tokens, $0.10 for cached input and $2.50 for cache writes. That is one fifth of GPT-6 Astra's $10 / $50. Requests over 272K input tokens cost 2x input and 1.5x output.
What is the GPT-6.1 Sol model ID?
gpt-6.1-sol. It has a 1,050,000-token context window, 128,000 max output tokens and a 30 April 2026 knowledge cutoff. Reasoning effort goes from low to max, with medium as the default.
Is GPT-6.1 Sol better than GPT-6 Astra?
Not overall. OpenAI says it matches Astra on DeepSWE coding at about a fifth of the cost, but Astra still leads on computer use and scientific research tasks. For coding and high-volume agent work Sol is the better value; for the hardest problems Astra remains the top model.
Originally published on Proje Defteri, where this post is kept up to date.
More: English posts and free browser tools.
Your support means a lot! ✨ Comment 💬, like 👍, and follow 🚀 for future posts!
Top comments (1)
The $0.10 cached input cost for GPT-6.1 Sol is a smart tweak-5% of standard input rate instead of 10%-but it only helps if your workflow has heavy caching. For most users, the real cost savings come from avoiding Astra's $50 output rate, especially in coding where Sol's DeepSWE score matches Astra at 1/5 the cost. Still, the 30 April 2026 cutoff means it won't handle recent tech shifts.