OpenAI releases GPT-6.1 Sol at a fifth of GPT-6 Astra’s token prices

OpenAI has released GPT-6.1 Sol, a cheaper model that it says nearly matches GPT-6 Astra on agentic coding, computer use and professional work. It costs a fifth of Astra’s standard input and output token prices, OpenAI said on Tuesday at its DevDay 2026 event.

The model is an upgrade to GPT-6 Sol, which OpenAI released on 22 September alongside a cheaper Luna model. The new model is available now to Plus, Pro, Business, Enterprise and Edu users in ChatGPT Work and Codex. It is not yet available in the main ChatGPT chat. Developers can use it through the API as gpt-6.1-sol.

“Near-Astra intelligence for a fifth of the price,” OpenAI wrote in its announcement.

In the API, GPT-6.1 Sol costs $2 per million input tokens and $10 per million output tokens. Cached input, which lets developers reuse context across requests, costs $0.10 per million tokens. That is 95% less than the standard input price and half of GPT-6 Sol’s cached input price, according to OpenAI.

OpenAI will also offer a faster version of Sol in its Ultrafast tier in the coming days. It says Ultrafast generates tokens up to eight times faster in Codex.

On DeepSWE v1.1, a test of software engineering tasks in real codebases, Sol matches Astra at roughly a fifth of the cost, OpenAI said. It beats GPT-6 Sol’s best score by 6.4 percentage points.

On AutomationBench, which tests multi-step business workflows, OpenAI said Sol scores 2.2 points above Anthropic’s Opus 5.5 at about a third of the cost. On the OSWorld 2.0 computer-use test, it comes within 2.1 points of Astra at about a seventh of the cost per task.

On Terminal-Bench Science 0.1, Sol costs $5.47 per task on average at maximum effort, according to OpenAI. That compares with $23.21 for Opus 5.5 and $23.80 for Astra. Astra still has the highest score of the models tested, at 68.1%. OpenAI said it should be used for the hardest scientific research tasks.

OpenAI ran its own models’ tests in its research environment or through its API. It took competitors’ results from public reports.

“GPT-6.1 Sol showed the kind of problem-solving we want from an AI coding partner. It helped identify important accessibility and language-support issues, and worked through limitations in our test setup to check the application more thoroughly,” said Jayesh Govindarajan, executive vice president of software engineering at Salesforce.

At low reasoning effort, Sol cut the share of answers with a factual error from 11.4% to 7.7%, OpenAI said. That test used deliberately difficult prompts on which users had flagged an earlier model’s mistake.

In one safety test, Sol failed to tell users that their search tool was broken in 2.1% of cases. The figure was 4.9% for GPT-6 Sol and 1.5% for Astra. OpenAI said it saw no attempts by Sol to bypass an automated safety reviewer. Details are in the model’s system card addendum .

OpenAI released GPT-6 Astra on 3 September. It has cancelled the October launch of GPT-6.1 Astra after failed safety tests. At the same event, OpenAI launched a $500 Pro plan and halved Pro 200 usage.