GPT-6.1 Sol: what OpenAI launched, what it costs and who can use it

GPT-6.1 Sol is OpenAI’s newest mid-priced model, announced at DevDay on September 29, 2026, a week after GPT-6 Sol and Luna. OpenAI says it “nearly matches GPT-6 Astra’s intelligence” on agentic coding, computer use and professional work, at one-fifth of Astra’s standard token prices. This report covers the price, what OpenAI claims, where you can use it and what isn’t there yet, using OpenAI’s launch post, its DevDay recap and its release notes.

GPT-6.1 Sol at a glance

The standard price is the same as GPT-6 Sol’s, so the upgrade is about capability and caching rather than a cut in headline price. Cached input now costs $0.10 per million tokens, which OpenAI says is 95% less than standard input pricing and 50% less than GPT-6 Sol’s cached price, and is aimed at agents that reuse context across requests.

What OpenAI says it can do

OpenAI’s post lists results across several benchmarks, all run by OpenAI in its research environment or API, which it warns can differ from production ChatGPT:

• Coding. On DeepSWE v1.1, a test of long software-engineering tasks in real code bases, it matches GPT-6 Astra at roughly one-fifth of the cost and beats GPT-6 Sol’s best score by 6.4 percentage points at lower reasoning effort and cost.
• Professional documents. On GDP.pdf, which asks professional questions about complex PDFs, OpenAI says it scores higher than Claude Opus 5.5 with fallbacks at less than half the cost per task.
• Business workflows. On AutomationBench it scores 2.2 percentage points above Opus 5.5 at medium effort at roughly a third of the cost, and 4.8 points above GPT-6 Sol.
• Computer use. On OSWorld 2.0’s offline set it beats GPT-6 Sol by seven percentage points at maximum effort at less than half the cost, and comes within 2.1 points of Astra at roughly one-seventh of the cost.
• Science. On Terminal-Bench Science it more than doubles GPT-6 Sol’s score at maximum effort. It costs $5.47 a task on average at maximum effort, against $23.21 for Opus 5.5 and $23.80 for Astra, but Astra still scores highest at 68.1% and OpenAI says to use it for the hardest research tasks.
• Factual accuracy. At low reasoning effort it reduces the share of responses containing a factual error on a deliberately difficult set of prompts from 11.4% to 7.7%, about 32% fewer. OpenAI notes those prompts aren’t typical usage.

Read these as a vendor’s claims. OpenAI’s own footnote says one Claude data point understates its cost because it leaves out fallbacks, which happened on about 40% of tasks.

Safety notes

OpenAI says GPT-6.1 Sol is closer to Astra than GPT-6 Sol on its alignment tests. On a test of whether agents tell you when their search tool is broken instead of guessing, GPT-6.1 Sol fails to disclose the problem in 2.1% of cases, against 4.9% for GPT-6 Sol, 1.5% for Astra and 28.7% for GPT-6 Luna, on tasks chosen to provoke failures. It says it saw no attempts to bypass an automated safety reviewer. Full details are in its system card addendum.

Who can use it

OpenAI says GPT-6.1 Sol is available from September 29 to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex, and that it is not yet available in Chat. Its release notes say the rollout starts with Pro users and expands to the other plans. On Enterprise and Edu plans, workspace owners can enable model access in Workspace settings under Permissions and roles, including through a custom role. Developers can call gpt-6.1-sol in the API.

Our guide to ChatGPT models explained shows which model each plan uses, and What is ChatGPT Work? explains the agent these models power.

Ultrafast speed tier

OpenAI also introduced Ultrafast, a premium speed tier with up to 8x faster token generation (300 tokens per second) in Codex and up to 6x in the API. GPT-6 Astra Ultrafast is available now in the API and in ChatGPT Work and Codex on the new Pro 500 plan and on Enterprise. GPT-6.1 Sol Ultrafast is coming in the next few days. See ChatGPT Pro explained for the Pro tiers.

How it compares with Claude’s latest

The launch lands in the same week as Anthropic’s Claude Sonnet 5.5, priced at $2 and $10 per million tokens as well, and a week after Claude Opus 5.5 at $4 and $20. The two vendors publish different benchmark sets, so the numbers are not directly comparable, and independent testing will say more than either launch post. For the coding tools that use these models, see the best AI coding assistants and the rest of the week’s news in OpenAI DevDay 2026: everything announced.