❦The Daily Commit❦ ☾The Nightly Build☽

OpenAI has released GPT-6.1 Sol as an upgrade to GPT-6 Sol. The model targets agentic coding and complex professional work. OpenAI says it approaches GPT-6 Astra on demanding tasks while costing about one-fifth as much at standard input and output rates. GPT-6 Astra remains the company’s strongest frontier model.

The API price is $2 per million input tokens, $0.10 per million cached input tokens and $10 per million output tokens. Cached input is priced 95% below the standard input rate, which reduces the cost of workflows that reuse context across requests.

On DeepSWE v1.1, a benchmark for software work in real codebases, GPT-6.1 Sol reaches GPT-6 Astra’s reported performance at roughly one-fifth of the cost. It leads the best GPT-6 Sol result by 6.4 percentage points while using less reasoning effort and costing less.

On GDP.pdf, which tests answers based on complex PDFs with tables, graphics, charts and fine print, GPT-6.1 Sol scores higher than Opus 5.5 with fallbacks across the tested reasoning settings. It costs less than half as much per task and comes close to GPT-6 Astra at about one-fifth of the cost. On AutomationBench, the model leads Opus 5.5 by 2.2 percentage points at the medium reasoning setting and costs about one-third as much. It also improves on GPT-6 Sol by 4.8 points at that setting.

AutomationBench 1.0.6 tests 47 tools spanning sales, marketing, operations, support, finance and HR in end-to-end workflows. OpenAI notes the Claude Fable 5.1 figure omits fallback costs, even though fallbacks occurred on about 40% of tasks.

GPT-6.1 Sol also raises computer-use results. On the offline OSWorld 2.0 dataset, it beats GPT-6 Sol by seven points at the maximum reasoning setting for less than half the cost. It trails GPT-6 Astra by 2.1 points while costing about one-seventh as much per task.

On Terminal-Bench Science 0.1, which covers scientific workflows such as data analysis, simulations and theorem proving, GPT-6.1 Sol scores more than twice as high as GPT-6 Sol at maximum reasoning. The average cost is $5.47 per task, compared with $23.21 for Opus 5.5 and $23.80 for GPT-6 Astra. GPT-6 Astra still has the top score at 68.1% and remains the recommended choice for the hardest research tasks.

The model’s factual error rate is 4.1% at extra-high reasoning, compared with 4.5% for GPT-6 Sol and 4.0% for GPT-6 Astra. Its cost per task is about 83% below Astra’s. This evaluation uses de-identified conversations in which users had reported an earlier model error. The prompts were selected to be difficult and do not represent typical use.

OpenAI also reports better alignment than GPT-6 Sol. In difficult tests, GPT-6.1 Sol more reliably discloses broken search tools and avoids unauthorized results in agentic tasks. OpenAI observed no attempts to bypass automated safety checks, as with GPT-6 Astra and GPT-6 Sol. In the broken-search-tool test, the model failed to flag the problem in 2.8% of cases, compared with 4.9% for GPT-6 Sol, 1.5% for GPT-6 Astra and 28.7% for GPT-6 Luna. The test used maximum reasoning and was designed to provoke errors.

GPT-6.1 Sol is now available to Plus, Pro, Business, Enterprise and Edu subscribers in ChatGPT Work and Codex. It is not available in Chat. Developers can call it through the API with the model ID gpt-6.1-sol.

OpenAI has also introduced GPT-6 Astra Ultrafast and GPT-6.1 Sol Ultrafast. Astra Ultrafast is described as the fastest frontier model worldwide, with up to eight times the speed of GPT-6 Astra. Both Ultrafast models are available through the API and in ChatGPT Work and Codex with the new Pro tier, which costs $500 per month and provides the highest usage limits. GPT-6.1 Sol Ultrafast targets near-Astra capability at up to eight times the speed for roughly the previous GPT-6 Astra API cost.