GPT-6.1 Sol
GPT-6.1 Sol is the upgrade to GPT-6 Sol, released by OpenAI on Sep 29, 2026. It costs US$ 2 per million input tokens and US$ 10 per million output, one fifth of GPT-6 Astra, and rereads the cache for US$ 0.10, half of GPT-6 Sol. In OpenAI's charts it matches Astra on coding and comes close on computer use and documents, but falls well behind on science. It is the model OpenAI recommends for complex work where cost matters.
GPT-6.1 Sol in official numbers
Data published by the maker in its documentation and announcement. The source of each item is at the end of the page.
- Maker
- OpenAI
- API identifier
gpt-6.1-sol- Release
- Sep 29, 2026, on DevDay 2026
- Input
- US$ 2 per million tokens
- Output
- US$ 10 per million tokens
- Cache read
- US$ 0.10 per million (5% of input)
- Cache write
- US$ 2.50 per million (1.25x input)
- Above 272 thousand input tokens
- input and cache doubled and output 1.5x, for the whole request: US$ 4 and US$ 15
- Batch and Flex
- US$ 1 input and US$ 5 output
- Fast mode
- US$ 4 and US$ 20; Ultrafast has been announced and has no price yet
- Context window
- 1,050,000 tokens (up to 922 thousand input)
- Max output
- 128 thousand tokens
- Knowledge cutoff
- Apr 30, 2026
- Input and output
- text and image in, text out; no audio or video
- Reasoning
- effort
lowtomax, defaultmedium; nononeorminimal - Where it runs
- OpenAI API (Responses, Batch and Chat Completions without tools), ChatGPT Work and Codex
- Data residency
- US and EU; no Fast mode with EU residency
- Cyber risk
- Critical on the Preparedness scale, with the same safeguards as Astra
Where GPT-6.1 Sol does well, where it falls behind and who it is for
Where it does well
- Coding: 75.2% on DeepSWE at high effort, for US$ 0.65 per task, in OpenAI's charts. Astra's best is 74.1%, for US$ 4.43.
- Computer use: 71.4% on OSWorld 2.0 offline at max effort, 2.1 points below Astra, at about one seventh of the cost per task (US$ 1.27 against US$ 9.44).
- Science for the money: 57.0% on Terminal-Bench Science for US$ 5.47 per task, against US$ 23.21 for Claude Opus 5.5 and US$ 23.80 for Astra in the same chart.
- Professional documents: 32.0% on GDP.pdf, against 28.8% for the best Opus 5.5 result with fallback, at less than half the cost per task.
- Cheap cache: US$ 0.10 per million on rereads, half of GPT-6 Sol and Claude Sonnet 5.5.
Where it falls behind
- Hard science: on Terminal-Bench Science, Astra scores 68.1% and Opus 5.5 63.3%, against 57.0% for 6.1 Sol. OpenAI itself says to use Astra for the hardest scientific tasks.
- Automation at max effort: on AutomationBench, Opus 5.5 with fallback reaches 42.5% at max effort, against 36.1% for 6.1 Sol. The announcement text only compares the two at medium effort.
- Long context: above 272 thousand input tokens, it pays US$ 4 input and US$ 15 output for the whole request. Sonnet 5.5 charges US$ 2 and US$ 10 up to 1 million.
- Fewer API options: no
noneeffort, no tools on Chat Completions, no fine-tuning and no predicted outputs. GPT-6 Sol acceptsnoneand tools on Chat Completions. - Agent behavior: in the safety report, it pushed past warnings in 23.5% of runs, against 17.4% for Astra, and was slightly more deceptive in coding than GPT-6 Sol (1.50% against 1.30%).
Who it makes sense for
- Complex projects where cost matters, like OpenAI's own examples: a board presentation built from financial results, or a website built from a product brief.
- Long-running coding agents, in Codex or on the API: it is the model OpenAI recommends for complex coding and agentic workflows.
- Apps with long, repeated instructions, where cache at US$ 0.10 per million cuts the bill.
- Current GPT-6 Sol users who do not rely on
none: same input and output price, half the cache price.
When to pick another model
- The hardest scientific research, where Astra scores 68.1% against 57.0% on Terminal-Bench Science.
- Integrations that call tools through Chat Completions or use
noneeffort: stay on GPT-6 Sol or Luna. - Requests that often go past 272 thousand tokens: Claude Sonnet 5.5 keeps US$ 2 and US$ 10 up to 1 million.
- High-volume classification, extraction and summaries: GPT-6 Luna costs US$ 0.10 and US$ 0.50 per million tokens.
What OpenAI published at launch
6.1 Sol came out on Sep 29, 2026, on DevDay, one week after GPT-6 Sol. OpenAI sums it up as near-Astra intelligence for a fifth of the price. At the same input and output price as GPT-6 Sol, it gets half-price cache and beats its predecessor on all six performance tests in the announcement.
The pattern repeats: 6.1 Sol stays close to Astra while spending a fraction, and Astra only pulls away in science. Against Claude Opus 5.5, which appears in the charts with fallback, 6.1 Sol wins on documents and loses on automation and science.
One caveat: Astra shows 68.1% on Terminal-Bench Science and 73.5% on OSWorld here, against 64.6% and 72.6% in its own launch announcement.
| GPT-6.1 Sol OpenAI | GPT-6 Astra OpenAI | GPT-6 Sol OpenAI | Claude Opus 5.5 Anthropic, with fallback | |
|---|---|---|---|---|
| DeepSWE v1.1 software engineering |
75.2% (US$ 0.65) | 74.1% (US$ 4.43) | 68.8% (US$ 2.74) | not measured |
| GDP.pdf professional PDF documents |
32.0% (US$ 0.35) | 32.2% (US$ 1.91) | 28.0% (US$ 0.35) | 28.8% (US$ 0.83) |
| AutomationBench automation |
36.1% (US$ 0.30) | 41.4% (US$ 1.73) | 33.2% (US$ 0.27) | 42.5% (US$ 1.44) |
| OSWorld 2.0 offline computer use |
71.4% (US$ 1.27) | 73.5% (US$ 9.44) | 64.4% (US$ 3.37) | not measured |
| Terminal-Bench Science 0.1 science in the terminal |
57.0% (US$ 5.47) | 68.1% (US$ 23.80) | 27.6% (US$ 12.18) | 63.3% (US$ 23.21) |
| Factual errors hard prompts, lower is better |
4.1% (US$ 0.10) | 3.9% (US$ 0.48) | 4.5% (US$ 0.13) | not measured |
Charts from the GPT-6.1 Sol announcement. Each model's best score at any effort level; in brackets, the average cost per task. The dot marks the best in each row. Competitor scores come from public reports, according to OpenAI.
GPT-6.1 Sol against Claude Sonnet 5.5 and Opus 5.5: price
The direct price rival is Claude Sonnet 5.5: the same US$ 2 input, US$ 10 output and US$ 2.50 cache write. 6.1 Sol wins on cache rereads, US$ 0.10 against US$ 0.20, which matters for apps with long, repeated instructions.
The math flips above 272 thousand input tokens. There OpenAI doubles input and cache and charges 1.5x output for the whole request, while Anthropic keeps the same rates up to 1 million. On a request that size, 6.1 Sol pays twice Sonnet 5.5's input price and ties on cache.
Against Claude Opus 5.5, which costs twice as much per token, OpenAI's charts put 6.1 Sol ahead on documents, at less than half the cost, and on automation at medium effort, at about a third. At max effort, Opus 5.5 pulls ahead on automation and science.
| GPT-6.1 Sol OpenAI | Sonnet 5.5 Anthropic | Opus 5.5 Anthropic | |
|---|---|---|---|
| Input | US$ 2 | US$ 2 | US$ 4 |
| Output | US$ 10 | US$ 10 | US$ 20 |
| Cache read | US$ 0.10 | US$ 0.20 | US$ 0.20 |
| Cache write | US$ 2.50 | US$ 2.50 | US$ 5 |
| Input above 272 thousand tokens | US$ 4 | US$ 2 | US$ 4 |
| Output above 272 thousand tokens | US$ 15 | US$ 10 | US$ 20 |
| Cache above 272 thousand tokens | US$ 0.20 | US$ 0.20 | US$ 0.20 |
| Context window | 1.05 million | 1 million | 1 million |
OpenAI and Anthropic pricing pages, checked on Oct 1, 2026. For Anthropic, 5-minute cache writes.
Astra, 6.1 Sol, Sol or Luna: choosing within OpenAI
6.1 Sol sits in the middle of the family. Astra costs five times as much per token and, by OpenAI's own charts, only pays off in science and the most demanding work. GPT-6 Luna costs one twentieth and suits clear, repeatable tasks.
The original GPT-6 Sol remains on the API at the same input and output price, but its page already points to 6.1 Sol as the newer Sol model. It still makes sense for integrations that depend on none effort or on tools through Chat Completions.
| GPT-6 Astra | GPT-6.1 Sol | GPT-6 Sol | GPT-6 Luna | |
|---|---|---|---|---|
| Input | US$ 10 | US$ 2 | US$ 2 | US$ 0.10 |
| Output | US$ 50 | US$ 10 | US$ 10 | US$ 0.50 |
| Cache read | US$ 1 | US$ 0.10 | US$ 0.20 | US$ 0.01 |
| Cache write | US$ 12.50 | US$ 2.50 | US$ 2.50 | US$ 0.125 |
| Above 272 thousand tokens input and output |
US$ 20 and US$ 75 | US$ 4 and US$ 15 | US$ 4 and US$ 15 | US$ 0.20 and US$ 0.75 |
| Context window | 1.05 million | 1.05 million | 1.05 million | 1.05 million |
| Knowledge cutoff | Apr 30, 2026 | Apr 30, 2026 | Apr 20, 2026 | May 18, 2026 |
| Accepts effort none | no | no | yes | yes |
| Role, per OpenAI | highest intelligence | balance of cost and intelligence | superseded by 6.1 Sol | fastest and most cost-effective |
OpenAI documentation, pricing page and model pages, checked on Oct 1, 2026.
What changes when coming from GPT-6 Sol
OpenAI asks anyone using gpt-6-sol to read the migration guide before switching. The differences that can break an integration or change the bill:
- No
noneand nominimaleffort. The default is stillmedium. - Tools only through the Responses API. Chat Completions works, but without tool calling.
- Cache reads drop from US$ 0.20 to US$ 0.10 per million. Input, output and cache writes stay the same.
- Knowledge cutoff of Apr 30, 2026, against Apr 20, 2026 for GPT-6 Sol.
- It moves to Critical in cybersecurity and inherits Astra's safeguards, including misalignment monitoring, which can stop a task.
Examples of using GPT-6.1 Sol
Financial statements, spreadsheets and management comments become a board presentation, revised over several rounds. It is one of the examples OpenAI gives for 6.1 Sol, and GDP.pdf measures how well a model reads this kind of document.
The agent fixes bugs, writes tests and proposes changes throughout the day. On DeepSWE, 6.1 Sol at high effort scored 75.2% for US$ 0.65 per task, where Astra needed US$ 4.43 to reach 74.1%.
The company handbook, catalog and policies go in as a fixed instruction and stay cached. Every question rereads that block for US$ 0.10 per million tokens. Keep requests under 272 thousand tokens to stay in the lower price band.
Where GPT-6.1 Sol stands out and where it falls behind, by the numbers
GPT-6.1 Sol has published results on 23 tests compiled by Epoch AI. The yardstick here is its generation: its position among the models released between Jan 2, 2026 and Dec 28, 2026 that took the same test. Against the whole archive, which includes 2023 models, a recent release lands near the top of almost everything, and that says nothing.
Where it stands out
Tests where it sits in the top third of its own generation.
Where it falls behind
Tests where it sits in the bottom half of its own generation.
It is not in the bottom half of its generation on any test with 6 or more models compared.
See all 23 tests, one by one
| Test | Score | In its generation | Overall | Test record |
|---|---|---|---|---|
| GPQA Diamond | 95.4% | #4 / 73 (tied with 1) | #4 / 173 | 96.4% Gemini 4 Argon |
| AIME 2024 and 2025 (OTIS) | 100.0% | #1 / 75 (tied with 11) | #1 / 168 | 100.0% GPT-5.5 |
| CritPt | 31.7% | #2 / 76 (tied with 2) | #2 / 112 | 32.3% GPT-5.6 Sol |
| SciCode | 55.7% | #25 / 74 | #25 / 104 | 66.9% Claude Opus 5.5 |
| Frontiermath Tiers 1 3 V2 | 93.7% | #1 / 64 (tied with 1) | #1 / 86 | 93.7% GPT-6 Astra |
| ARC-AGI 2 | 94.2% | #2 / 43 | #2 / 83 | 95.0% GPT-6 Astra |
| ALE-Bench | 2,744 | #2 / 57 | #2 / 82 | 2,951 GPT-6 Astra |
| WebDev Arena | 1,759 | #4 / 67 | #4 / 82 | 1,820 Claude Opus 5.5 |
| ARC-AGI | 98.5% | #1 / 43 (tied with 4) | #1 / 81 | 98.5% Claude Fable 5 |
| SimpleQA Verified | 73.9% | #3 / 59 | #3 / 81 | 76.0% Gemini 4 Argon |
| Chess puzzles | 68.0% | #2 / 58 (tied with 1) | #2 / 80 | 76.0% GPT-6 Astra |
| Game puzzles | 80.0% | #2 / 58 | #2 / 77 | 84.0% GPT-6 Astra |
| ProofBench | 99.0% | #4 / 64 (tied with 3) | #4 / 74 | 100.0% Claude Fable 5.1 |
| Frontiermath Tier 4 V2 | 100.0% | #1 / 54 | #1 / 67 | 100.0% GPT-6.1 Sol |
| Apex Agents | 60.0% | #12 / 48 | #12 / 50 | 82.4% Gemini 4 Argon |
| FrontierCode | 50.2% | #7 / 43 | #7 / 43 | 54.6% Claude Opus 5.5 |
| Gdp Pdf | 35.0% | #1 / 36 | #1 / 37 | 35.0% GPT-6.1 Sol |
| Furniture Assembly | 80.0% | #2 / 32 (tied with 1) | #2 / 34 | 83.3% Claude Opus 5.5 |
| Ebr Bench | 54.3% | #5 / 22 | #5 / 27 | 76.2% GPT-6 Astra |
| Weirdml V3 | 38.3% | #2 / 10 | #2 / 11 | 42.2% GPT-6 Astra |
| Spireclimb | 24.7% | #2 / 7 | #2 / 7 | 25.7% GPT-6 Astra |
| Spireclimb Sts1 A20 Heart | 20.0% | #1 / 7 | #1 / 7 | 20.0% GPT-6.1 Sol |
| Spireclimb Sts2 A10 | 29.3% | #2 / 7 | #2 / 7 | 34.7% GPT-6 Astra |
For each test, the model's best effort level counts, as in the ranking. Source of the scores: Epoch AI, under CC BY 4.0.
What to compare GPT-6.1 Sol with
The models from other labs closest in capability and released in the same period, at most two per lab. Click any of them to see its page.
| Model | Lab | ECI | ELO | LiveBench | Release |
|---|---|---|---|---|---|
| Claude Opus 5.5 | Anthropic | 166.82 | 1,584 | 83.2 | Sep 22, 2026 |
| GPT-6.1 Sol | OpenAI | 166.32 | n/a | 81.6 | Sep 29, 2026 |
| Claude Fable 5.1 | Anthropic | 164.65 | 1,586 | 83.4 | Sep 1, 2026 |
| Gemini 4 Argon | Google DeepMind | 164.56 | 1,608 | n/a | Sep 30, 2026 |
| Kimi K3 | Moonshot | 157.42 | 1,528 | 79.2 | Jul 16, 2026 |
| Gemini 3.7 Flash | Google DeepMind | 157.22 | 1,551 | 78.8 | Aug 13, 2026 |
| Grok Build 0.1 | xAI | no ECI | n/a | 67.8 | May 29, 2026 |
GPT-6.1 Sol or Claude Opus 5.5?
Claude Opus 5.5 is the model from another lab closest to GPT-6.1 Sol in capability. Across the 22 tests both took, this is the score:
| Test | GPT-6.1 Sol | Claude Opus 5.5 |
|---|---|---|
| Spireclimb Sts1 A20 Heart | 20.0% | 0.0% |
| Spireclimb | 24.7% | 3.3% |
| Spireclimb Sts2 A10 | 29.3% | 6.7% |
| Ebr Bench | 54.3% | 71.4% |
| ALE-Bench | 2,744 | 2,147 |
| Apex Agents | 60.0% | 73.5% |
| SciCode | 55.7% | 66.9% |
| Gdp Pdf | 35.0% | 30.6% |
| Game puzzles | 80.0% | 71.0% |
| FrontierCode | 50.2% | 54.6% |
Unweighted score: each test counts one point, whether math or coding. The dot marks who did better on each row.
Other models from OpenAI
The models from the same lab closest in time, newest first.
| Model | Release | ECI | ELO | LiveBench | Access |
|---|---|---|---|---|---|
| GPT-6.1 Sol | Sep 29, 2026 | 166.32 | n/a | 81.6 | API access |
| GPT-6 Sol | Sep 22, 2026 | 162.55 | n/a | 79.3 | API access |
| GPT-6 Luna | Sep 22, 2026 | 156.31 | n/a | 72.0 | API access |
| GPT-6 Astra | Sep 3, 2026 | 166.41 | n/a | 82.2 | API access |
| GPT-5.6 Sol | Jul 9, 2026 | 161.73 | 1,527 | 81.1 | API access |
| GPT-5.6 Terra | Jul 9, 2026 | 159.43 | 1,515 | 77.9 | API access |
| GPT-5.6 Luna | Jul 9, 2026 | 156.33 | 1,478 | 73.6 | API access |
| GPT-5.5 Instant | May 5, 2026 | 142.45 | 1,456 | n/a | API access |
| GPT-5.5 Pro | Apr 23, 2026 | 161.97 | n/a | n/a | API access |
Have you used GPT-6.1 Sol? Rate it
Users' opinion, not measurement: the rating is kept apart from the ranking and does not move anyone's position. One vote per browser, no sign-up, and you can change your rating later.
What people ask about GPT-6.1 Sol
How much does GPT-6.1 Sol cost?
US$ 2 per million input tokens and US$ 10 per million output. Cache reads cost US$ 0.10 and cache writes US$ 2.50. Above 272 thousand input tokens, the whole request moves to US$ 4 input and US$ 15 output. Batch and Flex are half price, and Fast mode costs double.
What is the difference between GPT-6.1 Sol and GPT-6 Sol?
Input and output prices are the same, and 6.1 Sol rereads the cache at half the price. In OpenAI's charts it beats its predecessor on all six published performance tests, with the widest gap in science in the terminal (57.0% against 27.6%). In exchange, it drops the none effort and tools through Chat Completions.
GPT-6.1 Sol or Claude Sonnet 5.5?
Up to 272 thousand input tokens, the price per token is the same, and 6.1 Sol rereads the cache at half the price. Above that, OpenAI charges more while Anthropic keeps its rates up to 1 million. The independent scores are at the top of this page; to decide, run your own tasks on both and compare cost per task.
Where can I use GPT-6.1 Sol?
On the OpenAI API, with US and EU data residency, and in ChatGPT Work and Codex on the Plus, Pro, Business, Enterprise and Edu plans. On Enterprise and Edu, an admin has to turn it on. It is not in Chat yet, and OpenAI's Amazon Bedrock page does not list it yet.
Where does GPT-6.1 Sol rank among AI models?
In the ROO3 AI Benchmark, ordered by the Epoch AI ECI, GPT-6.1 Sol ranks #3 out of 267 models, with an ECI of 166.32. The ECI is a composite index that combines several hard benchmarks into a single scale. The position changes when a new model comes out, and the page is refreshed six times a day.
Is GPT-6.1 Sol better than Claude Opus 5.5?
Across the 22 tests both models took and Epoch AI compiled, GPT-6.1 Sol comes out ahead on 12, behind on 7 and tied on 3. It is a test-by-test count with no weighting: a win in math counts the same as a win in coding. To choose between the two, look at the tests that resemble your own use.
What is GPT-6.1 Sol best at?
Against models of the same generation (released up to 9 months before and up to 3 months after it), the best relative results of GPT-6.1 Sol are: Frontiermath Tier 4 V2 (#1 of 54); Gdp Pdf (#1 of 36); Spireclimb Sts1 A20 Heart (#1 of 7). The yardstick is its position among the models of that period that took the same test.
Where do the numbers on this page come from?
From the same three sources as the ranking, all openly licensed: Epoch AI (ECI and individual tests, CC BY 4.0), Arena (human-preference ELO, CC BY 4.0) and LiveBench (overall average, CC BY-SA 4.0). ROO3 does not measure any model. When the page shows a figure published by the maker itself, it appears in the written analysis, labeled as the maker's, and never enters the ranking.
Where each number comes from
- OpenAI, GPT-6.1 Sol announcement
- OpenAI, GPT-6.1 Sol model page (documentation)
- OpenAI, API pricing
- OpenAI, Addendum to the GPT-6 Astra System Card: GPT-6.1 Sol
- OpenAI, latest model guide (family and migration)
- OpenAI, model selection guide
- OpenAI, models in Codex and ChatGPT Work
- OpenAI, API changelog
- OpenAI, models on Amazon Bedrock (documentation)
- Anthropic, pricing (documentation)
Epoch AI, "AI Benchmarking Hub". Published at epoch.ai. Data under the CC BY 4.0 licence, reordered by roo3.co.
The Arena (LMArena) leaderboard, from the public lmarena-ai/leaderboard-dataset on Hugging Face, under the CC BY 4.0 licence. We are not affiliated with the Arena.
LiveBench, by the LiveBench team, published at livebench.ai under the CC BY-SA 4.0 license. The overall average is recomputed by roo3.co with the same formula as the website (average of the categories), and those numbers stay under CC BY-SA 4.0. We are not affiliated with LiveBench.
← See the full ranking