OpenAI says GPT-6.1 Sol matched GPT-6 Astra on DeepSWE v1.1, a benchmark of AI agents tackling long software-engineering tasks in real codebases, at roughly one-fifth the task cost. The release does not show a task-level cost breakdown, so readers cannot reconstruct that figure from the published details.
OpenAI also says GPT-6.1 Sol scored 6.4 percentage points above GPT-6 Sol’s best DeepSWE result while using less reasoning effort and cost. The public DeepSWE leaderboard was last updated on September 22 and did not list GPT-6.1 Sol as of September 30, so it does not yet independently confirm the new comparison.
OpenAI notes that its model evaluations run in a research environment or through its API and can differ from production ChatGPT because prompts, tools and reasoning effort vary.
GPT-6.1 Sol is available through the API and in ChatGPT Work and Codex, though OpenAI says it is not yet available in ChatGPT. Its standard API prices are $2 per million input tokens, $0.10 for cached input and $10 per million output tokens. OpenAI lists its standard input and output token rates at one-fifth of GPT-6 Astra’s.
Sources
- OpenAI: Introducing GPT-6.1 Sol, benchmark results, evaluation caveats, prices and availability.
- DeepSWE, benchmark description and public leaderboard.
- Illustrative photo by Rob Wingate on Unsplash, free under the Unsplash License.