GPT-6.1 Sol review

Replaced GPT-6 Sol a week after launch, at a lower price.

Out nowOpenAIAI modelUpdated Oct 26, 2026
Try GPT-6.1 SolOpen in ChatGPT · official source, chatgpt.com
Released29 Sep 2026
Price (API)$2 in / $10 out
Best forCoding agents
First lookOur take, based on published facts and early independent tests

A cheaper way to get close to OpenAI's top model. OpenAI says it matches GPT-6 Astra on long coding tasks at about a fifth of the cost and makes fewer factual mistakes than GPT-6 Sol. Most numbers so far come from OpenAI itself.

What is GPT-6.1 Sol?

GPT-6.1 Sol is an OpenAI model released on 29 September 2026 at OpenAI DevDay, one week after GPT-6 Sol. It sits below GPT-6 Astra, OpenAI's flagship, and is built for coding agents, using a computer, long documents and business automation.

Key features

  • Close to GPT-6 Astra on coding and computer use, at a much lower price
  • Fewer factual errors: 7.7% against 11.4% for GPT-6 Sol on OpenAI's tests
  • Reads about 1 million tokens at once and writes up to 128,000
  • Five thinking levels, from low to max
  • In Codex and ChatGPT Work on paid plans, the OpenAI API and GitHub Copilot

Specs and price

Made byOpenAI
Released29 Sep 2026 (DevDay)
API price$2 per 1M input tokens, $10 per 1M output tokens
Repeated input (cache)$0.10 per 1M tokens
Very long promptsOver 272,000 tokens: 2x input price and 1.5x output price
How much it can read1,050,000 tokens (about 780,000 words)
Longest answer128,000 tokens
WhereCodex and ChatGPT Work on Plus, Pro, Business, Enterprise and Edu. Not in regular chat, Free or Go
Free to downloadNo
Name in the APIgpt-6.1-sol

Test scores

TestScorePublished by
DeepSWE v1.1Long, real coding tasksMatches GPT-6 Astra at about 1/5 of the costOpenAI
AutomationBenchAutomating business workflows2.2 points above Claude Opus 5.5OpenAI
SEC-Bench ProSecurity tasks78.8% (GPT-6 Astra: 85.4%)OpenAI
HealthBench ProfessionalMedical questions64.2 (GPT-6 Astra: 64.7)OpenAI
Terminal-Bench Science 0.1Cost per science coding task$5.47 (Opus 5.5: $23.21)OpenAI

Scores published by OpenAI are the maker's own numbers. We will add our own results in the hands-on review.

How to use GPT-6.1 Sol

  1. In Codex or ChatGPT Work: sign in on a Plus, Pro, Business, Enterprise or Edu plan and pick GPT-6.1 Sol. Enterprise and Edu admins have to turn it on first.
  2. For developers: use gpt-6.1-sol in the OpenAI API. Tool calling needs the Responses API.
  3. Also available in GitHub Copilot (Pro+, Max, Business, Enterprise) and on OpenRouter.

Who it is for

Good fit

  • Developers running coding agents who want near-Astra results for less
  • Teams automating office and document work
  • Paid ChatGPT users who work in Codex

Skip it if

  • Free ChatGPT users: it is not included
  • People who need the very best: GPT-6 Astra still scores higher on several tests

Pros and cons

Pros

  • Near-flagship results at $2 / $10 per million tokens
  • Fewer factual mistakes than GPT-6 Sol
  • Huge 1M token memory
  • Available almost everywhere: API, Codex, Copilot, OpenRouter

Cons

  • Not in regular ChatGPT chat or on free plans
  • Prompts over 272K tokens cost a lot more
  • Almost all scores come from OpenAI, with no independent re-tests yet
  • Tools only work through the Responses API

How it compares

GPT-6.1 Sol costs the same per token as Gemini 4 Argon's launch offer ($2 / $10) and half as much as Claude Opus 5.5 ($4 / $20). OpenAI puts it just above Opus 5.5 on AutomationBench, and says GPT-6 Astra still beats it on security and troubleshooting tests.

Hands-on review In testing

We are running GPT-6.1 Sol on real tasks. Results will be added to this page, at the same address.

Real tasksCoding, writing, research, agents
SpeedHow fast it answers in practice
Cost per taskWhat a typical job costs
VerdictWho should switch, who should wait

Related models