GPT-6.1 Sol
- Provider
- OpenAI
- Status
- Current
- Context
- 1,050,000 tok
- Price
- $2 / $10 /MTok
- Knowledge
- 2026-04
GPT-6.1 Sol is OpenAI’s mid-weight model, released at DevDay on 29 September 2026, a week after GPT-6 Sol (TechCrunch). It keeps Sol’s $2/$10 per million tokens, a fifth of GPT-6 Astra’s price, and closes most of the gap to Astra on capability.
Artificial Analysis scored it 52 on the Intelligence Index v4.3 at max effort, four points above GPT-6 Sol and one below Astra’s 53, at $0.72 per completed index task against Astra’s $3.26. No SWE-bench figure appears in the launch coverage.
Quick specs
| GPT-6.1 Sol | |
|---|---|
| Released | 29 September 2026 |
| Input price | $2.00 / MTok |
| Cached input | $0.10 / MTok (95% off) |
| Output price | $10.00 / MTok |
| Over 272K input tokens | 2x input, 1.5x output |
| Context window | 1,050,000 tokens |
| Max output | 128,000 tokens |
| Knowledge cutoff | 30 April 2026 |
| AA Intelligence Index | 52 (max), 50 (high), 48 (medium) |
| Cost per AA index task | $0.72 |
| API model ID | gpt-6.1-sol |
What changed from GPT-6 Sol
GPT-6 Sol was a price move with a one-point capability gain. GPT-6.1 Sol is the capability move at the same price.
Intelligence. Up four points on the index, from 48 to 52. The largest gains on Artificial Analysis’s component evaluations are 12 points on Terminal-Bench 4.0, 8 on AA-Omniscience accuracy, 6 on GDP.pdf and 5 each on Humanity’s Last Exam and GDPval-AA v2.1, reversing GPT-6 Sol’s knowledge-work regression.
Cost per task. Down 31%, from $1.05 to $0.72, even though it spends 10–30% more output tokens than GPT-6 Sol.
Caching. The cached-input discount rises from 90% to 95%, to $0.10 per million tokens.
Factuality. Artificial Analysis measured the hallucination rate at max effort falling from 60% to 54%. OpenAI says its factual error rate on difficult prompts at low effort fell from 11.4% to 7.7%.
Benchmark performance
| Measure | GPT-6.1 Sol | GPT-6 Sol | GPT-6 Astra |
|---|---|---|---|
| Intelligence Index v4.3 (max) | 52 | 48 | 53 |
| Coding Agent Index (max) | 60 | 57 | 62 |
| Cost per index task | $0.72 | $1.05 | $3.26 |
| Hallucination rate (max) | 54% | 60% | n/a |
| Price (in/out per MTok) | $2 / $10 | $2 / $10 | $10 / $50 |
Source: Artificial Analysis, independent. Coding Agent Index figures follow from its reported gains (three points on GPT-6 Sol’s 57, two below Astra’s 62). At xhigh effort GPT-6.1 Sol scores three points higher on that index than at max, one above Astra.
OpenAI’s own figures, vendor-run: it says GPT-6.1 Sol matches Astra on DeepSWE v1.1 and comes within 2.1 points on OSWorld 2.0 computer use at roughly a seventh of the cost per task (The Next Web).
How it compares
vs GPT-6 Astra
GPT-6 Astra scores 53 on the index at $10/$50 and $3.26 per task. GPT-6.1 Sol is one point behind at a fifth of the token price and under a quarter of the cost per task. Astra remains ahead on the Coding Agent Index at max effort (62 to 60), and its Critical-rated cyber capability is not part of Sol.
vs Claude Sonnet 5.5
Claude Sonnet 5.5, released a day earlier, has the same $2/$10 token price and scores four points higher (56). At max effort it costs $7.60 per completed index task, about ten times GPT-6.1 Sol’s $0.72.
vs Claude Opus 5
Claude Opus 5 scores 51 on the index at $5/$25 and $5.86 per task. GPT-6.1 Sol edges it on the index at an eighth of the cost per task, but Opus 5 publishes 96.0% on SWE-bench Verified and 79.2% on SWE-bench Pro; GPT-6.1 Sol publishes neither.
Known limitations
No SWE-bench figure. No SWE-bench Pro or Verified result appears in the launch coverage, and no standardised harness has produced one.
Not in regular chat. At launch it is in ChatGPT Work and Codex, not ordinary ChatGPT conversations.
Safety testing was mixed. OpenAI’s system card addendum reports fewer factual errors but higher rates of attempting to bypass security restrictions than GPT-6 Astra in some test scenarios (The Next Web).
Hallucination rate is still high. Artificial Analysis measured 54% at max effort on AA-Omniscience, improved but high in absolute terms.
Frequently asked questions
When was GPT-6.1 Sol released?
29 September 2026, at OpenAI’s DevDay, on the API as gpt-6.1-sol and in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise and Edu users.
How much does GPT-6.1 Sol cost?
$2 per million input tokens and $10 per million output tokens, the same as GPT-6 Sol and a fifth of GPT-6 Astra. Cached input is $0.10 per million. Requests over 272K input tokens cost double on input and 1.5 times on output.
Is GPT-6.1 Sol better than GPT-6 Sol?
Yes, at the same price. Artificial Analysis scores it 52 against 48 on the Intelligence Index and three points higher on the Coding Agent Index, with cost per completed task down from $1.05 to $0.72.
Is GPT-6.1 Sol as good as GPT-6 Astra?
Close. It scores one point below Astra on Artificial Analysis’s index (52 to 53) and two below on the Coding Agent Index at max effort, and OpenAI says it matches Astra on DeepSWE v1.1. It costs under a quarter as much per completed task.
Where does GPT-6.1 Sol rank?
It enters our best AI models ranking at #6, directly below GPT-6 Astra, and takes the board’s best-value slot.
Written at launch, 30 September 2026. Artificial Analysis figures are independent. OpenAI’s launch page was not reachable at the time of writing, so specs and prices are taken from OpenAI’s API documentation, and vendor benchmark claims from launch coverage.