Updated

OpenAI
GPT-6 Sol
OpenAI's GPT-6 model between Luna and Astra, trained like Astra and able to reply with reasoning off.
What is GPT-6 Sol?
GPT-6 Sol is the middle model of OpenAI's GPT-6 series, released on September 22, 2026.
It sits below the flagship GPT-6 Astra and above GPT-6 Luna, and OpenAI trained it with methods similar to Astra's to bring Astra's gains in coding, computer use and professional work to a faster, cheaper model. OpenAI says it makes about half as many factual mistakes as its predecessor and gives clearer, slightly shorter answers. OpenAI's docs now point to GPT-6.1 Sol as the newer Sol model. In Super Agent it can reply with reasoning off or think up to max.
- Provider
- OpenAI
- Context window
- 1.05M tokens
- Input
- Text, Image
- Reasoning
- Off to max
- Max output
- 128K tokens
- Knowledge cutoff
- Apr 20, 2026
- Released
- Sep 22, 2026
How GPT-6 Sol performs
Published by OpenAI
Results OpenAI reported for GPT-6 Sol. Each lab runs its own setup, so compare these with other models only inside the same table.
| Benchmark | Score | Setting |
|---|---|---|
| Agents' Last Exam | 56.36% | max effort |
| DeepSWE | 68.81% | max effort |
| OSWorld 2.0 | 64.43% | offline set, max effort |
| FrontierCode | 49.27% | max effort |
| AutomationBench | 33.18% | extra high effort |
| Factual error rate on difficult prompts (lower is better) | 4.52% | extra high effort |
Source: openai.com
On LMArena
People compare two anonymous answers and vote for the better one. Ranked here against the other chat models you can pick in Super Agent.
Text
#20 of 27 in Super Agent
All text prompts, with style control on. #62 of 418 on the full leaderboard, from 14,707 votes.
- 1Claude Opus 5.51509
- 2Claude Fable 51503
- 3Claude Fable 5.11503
- 4Claude Opus 51491
- 5Gemini 3.7 Flash1490
- 20GPT-6 Sol1459
Coding
#15 of 27 in Super Agent
Text prompts that ask for code. #38 of 413 on the full leaderboard, from 3,371 votes.
- 1Claude Fable 51551
- 2Claude Opus 5.51548
- 3GPT-6 Astra1543
- 4Kimi K31539
- 5Claude Fable 5.11538
- 15GPT-6 Sol1519
WebDev
#7 of 27 in Super Agent
Building working web apps in Code Arena. #9 of 149 on the full leaderboard, from 4,961 votes.
- 1Claude Opus 5.51807
- 2GPT-6 Astra1775
- 3Claude Sonnet 5.51768
- 4GPT-6.1 Sol1749
- 5Claude Fable 5.11743
- 7GPT-6 Sol1682
Math
#21 of 27 in Super Agent
Text prompts with math problems. #79 of 404 on the full leaderboard, from 827 votes.
- 1Claude Fable 5.11529
- 2Claude Fable 51521
- 3Claude Opus 51517
- 4GPT-6.1 Sol1513
- 5Claude Opus 5.51512
- 21GPT-6 Sol1450
Hard prompts
#20 of 27 in Super Agent
Complex, multi-part and demanding text prompts. #58 of 418 on the full leaderboard, from 9,052 votes.
- 1Claude Opus 5.51535
- 2Claude Fable 51529
- 3Claude Fable 5.11523
- 4Kimi K31516
- 5Claude Opus 51515
- 20GPT-6 Sol1485
Creative writing
#21 of 27 in Super Agent
Stories, poems and other original writing. #65 of 416 on the full leaderboard, from 3,130 votes.
- 1Claude Opus 5.51520
- 2Claude Fable 51501
- 3Gemini 3.7 Flash1492
- 4Claude Fable 5.11487
- 5Gemini 3.1 Pro Preview1481
- 21GPT-6 Sol1435
Vision
#16 of 23 in Super Agent
Prompts that include images, with style control on. #39 of 167 on the full leaderboard, from 3,221 votes.
- 1Claude Fable 51309
- 2Gemini 3.7 Flash1295
- 3GPT-6.1 Sol1294
- 4Claude Opus 5.51293
- 5Claude Opus 51288
- 16GPT-6 Sol1266
Source: LMArena, leaderboard published . Scores move as new votes come in.
What you can do with GPT-6 Sol
Middle tier of the GPT-6 series, between Luna and Astra
About half the factual mistakes of its predecessor, according to OpenAI
Clearer, less jargon-heavy answers carried over from Astra
Reasoning can be switched off for instant replies
1.05M-token context that also takes images
How to prompt GPT-6 Sol
- Use reasoning off for rewrites and brainstorming, and medium or higher for plans.
- Describe the business goal along with the task so it can make sensible choices along the way.
- If you need its newer sibling's accuracy, try the same prompt on GPT-6.1 Sol.
Where GPT-6 Sol fits best
Professional writing and planning
Agent tasks with several steps
Business workflows such as briefs, reports and content calendars
Quick answers with reasoning off
How GPT-6 Sol compares
| Model | LMArena text | LMArena coding | Context window | Reasoning |
|---|---|---|---|---|
| GPT-6 Sol | 1459#62 | 1519#38 | 1.05M tokens | Off to max |
| DeepSeek V4 Pro | 1458#64 | 1501#79 | 1M tokens | Off, medium, high, extra high |
| Kimi K2.6 | 1461#59 | 1516#44 | 256K tokens | On or off |
| Claude Sonnet 5 | 1462#58 | 1520#37 | 1M tokens | Off to max |
| Grok 4.6 | 1456#66 | 1510#60 | 500K tokens | Low to extra high |
Good to know before you start
- OpenAI has released GPT-6.1 Sol as its upgrade
- Below GPT-6 Astra on the hardest end-to-end work
- Outputs text, so finished images and clips come from the generation model Super Agent calls
Start chatting with GPT-6 Sol
Chat with GPT-6 Sol in Super AgentSelect GPT-6 Sol
Open Super Agent on Flashloop and choose GPT-6 Sol in the model picker.
Enter your prompt
Ask for what you need: an answer, a plan, or the images and videos you want the agent to make.
Send and keep going
Send the message. GPT-6 Sol answers in Super Agent and runs any generations you asked for.
Questions about GPT-6 Sol
Where does GPT-6 Sol sit in the GPT-6 lineup?
Between GPT-6 Luna, the efficient low-cost tier, and GPT-6 Astra, OpenAI's flagship.
Should I use GPT-6 Sol or GPT-6.1 Sol?
GPT-6.1 Sol is OpenAI's upgrade, with higher coding and computer-use scores and fewer factual errors. GPT-6 Sol is still useful when you want to turn reasoning off, which GPT-6.1 Sol does not support.
What reasoning levels does GPT-6 Sol support?
Off, low, medium, high, extra high and max. Super Agent starts it at low.
What is GPT-6 Sol Pro?
GPT-6 Sol run in OpenAI's pro reasoning mode, which spends more work on each request. It sits directly under GPT-6 Sol in the model picker.
More chat models on Flashloop
DeepSeek V4 ProDeepSeek's largest V4 model for coding, knowledge and STEM; Super Agent runs the April preview.
Kimi K2.6Moonshot AI's open coding model that turns prompts and visual inputs into working interfaces.
Claude Sonnet 5Anthropic's most agentic Sonnet at launch, close to Opus 4.8 and able to reply with thinking off.
Grok 4.6xAI's Grok 4.6 for turning a broad product idea into a working first version, with 500K context.