Updated

A polished brushed-aluminium sphere with one thin floating halo ring on a dark background, abstract hero artwork for GPT-6.1 Sol

OpenAI

GPT-6.1 Sol

OpenAI's GPT-6 Sol upgrade, close to GPT-6 Astra on agentic coding and computer use for far less.

Chat with GPT-6.1 Sol in Super AgentView pricing

What is GPT-6.1 Sol?

GPT-6.1 Sol is OpenAI's upgrade to GPT-6 Sol, announced on September 29, 2026.

OpenAI says it nearly matches the flagship GPT-6 Astra on agentic coding, computer use and professional work at one-fifth of Astra's standard token rates, and recommends it for complex coding, research and computer use. It also gets fewer facts wrong: on OpenAI's set of difficult prompts, the share of factually wrong answers at low effort fell from 11.4% on GPT-6 Sol to 7.7%. Thinking is always on, with effort from low to max in Super Agent.

Provider
OpenAI
Context window
1.05M tokens
Input
Text, Image
Reasoning
Low to max
Max output
128K tokens
Knowledge cutoff
Apr 30, 2026
Released
Sep 29, 2026

How GPT-6.1 Sol performs

Published by OpenAI

Results OpenAI reported for GPT-6.1 Sol. Each lab runs its own setup, so compare these with other models only inside the same table.

Benchmark results OpenAI published for GPT-6.1 Sol
BenchmarkScoreSetting
DeepSWE75.22%high effort
OSWorld 2.071.42%offline set, max effort
Terminal-Bench Science 0.157.02%max effort
AutomationBench36.1%max effort
GDP.pdf32.0%high effort
Factual error rate on difficult prompts (lower is better)4.12%extra high effort

Source: openai.com

What you can do with GPT-6.1 Sol

  • Matches GPT-6 Astra on DeepSWE v1.1 at about a fifth of the cost, OpenAI reports

  • Within 2.1 points of Astra on OSWorld 2.0 at max effort

  • Fewer factually wrong answers on hard prompts than GPT-6 Sol at most effort levels, with the biggest drop at low effort

  • 1.05M-token context window with image input

  • Effort from low to max, starting at low in Super Agent

How to prompt GPT-6.1 Sol

  • Attach the source document instead of pasting fragments, so it works from the whole thing.
  • Start at low effort and raise it only when a plan or answer comes back too shallow.
  • Ask it to mark any claim it isn't sure about. Factual accuracy is one of the gains OpenAI measured in this release.

Where GPT-6.1 Sol fits best

  • Agentic coding and multi-step agent runs

  • Turning long documents into scripts, shot lists or briefs

  • Research that has to hold up to fact-checking

  • Everyday work that needs fewer factual slips

How GPT-6.1 Sol compares

GPT-6.1 Sol compared with other chat models on Flashloop
ModelLMArena textLMArena codingContext windowReasoning
GPT-6.1 Sol1480#271537#141.05M tokensLow to max
GLM-5.31481#261521#331M tokensLow, medium, high, max
Claude Opus 4.81482#251533#181M tokensOff to max
Claude Sonnet 5.51478#291528#261M tokensLow to max
Gemini 3.5 Flash1477#301508#621M tokensLow, medium, high

Good to know before you start

  • Thinking can't be turned off; OpenAI's API has no none setting for this model
  • No image or video output of its own: it writes the prompts, and Super Agent hands them to Flashloop's generation models
  • GPT-6 Astra still scores higher on some of OpenAI's own charts, such as Terminal-Bench Science

Start chatting with GPT-6.1 Sol

Chat with GPT-6.1 Sol in Super Agent

  1. Select GPT-6.1 Sol

    Open Super Agent on Flashloop and choose GPT-6.1 Sol in the model picker.

  2. Enter your prompt

    Ask for what you need: an answer, a plan, or the images and videos you want the agent to make.

  3. Send and keep going

    Send the message. GPT-6.1 Sol answers in Super Agent and runs any generations you asked for.

Questions about GPT-6.1 Sol

What changed from GPT-6 Sol to GPT-6.1 Sol?

OpenAI says GPT-6.1 Sol scores higher on agentic coding and computer use and makes fewer factual mistakes. On DeepSWE v1.1 it beats GPT-6 Sol's best score by 6.4 points, and its rate of factually wrong answers on difficult prompts at low effort dropped from 11.4% to 7.7%.

How does GPT-6.1 Sol compare with GPT-6 Astra?

Astra is still OpenAI's flagship. OpenAI reports that GPT-6.1 Sol matches Astra on DeepSWE v1.1 and comes within 2.1 points of it on OSWorld 2.0 at max effort, at a fraction of the cost per task.

Can I turn reasoning off with GPT-6.1 Sol?

No. OpenAI's API doesn't offer a none setting for GPT-6.1 Sol. In Super Agent you choose effort from low to max, and it starts at low.

What is the difference between GPT-6.1 Sol and GPT-6.1 Sol Pro?

They run the same model. Pro uses OpenAI's pro reasoning mode, which does more work on each request, so it is slower and uses more tokens. OpenAI describes pro as a mode for difficult tasks that can tolerate the extra latency.

Can GPT-6.1 Sol make images or videos in Super Agent?

The model itself writes text. In Super Agent it plans the work and writes the prompts, and the agent creates images and videos with Flashloop's generation models.

More chat models on Flashloop

Browse all models