OpenAI Launches GPT-6.1 Sol: Pricing, Benchmarks, and Availability

OpenAI introduced GPT-6.1 Sol on September 29, 2026, as an upgrade to GPT-6 Sol for coding, computer use, and professional work. The company says the model approaches GPT-6 Astra’s performance on several evaluations while costing less to run. The results below are reported by OpenAI, rather than a guarantee of performance in every workload.

API pricing GPT-6.1 Sol’s standard API prices are $2 per million input tokens, $0.10 per million cached input tokens, and $10 per million output tokens. For comparison, GPT-6 Astra’s standard input and output prices are $10 and $50 per million tokens. That makes Sol’s standard input and output rates one-fifth of Astra’s. Cached input is 95% cheaper than Sol’s uncached input and half the cached-input price of GPT-6 Sol.

Coding and professional-work results On DeepSWE v1.1, OpenAI reports that GPT-6.1 Sol matched Astra’s score at roughly one-fifth the cost per task. It exceeded GPT-6 Sol’s best score by 6.4 percentage points while using a lower reasoning-effort setting.

On GDP.pdf, a test involving questions about complex professional documents, Sol scored above Opus 5.5 with fallbacks at less than half its cost per task across the tested settings. It approached Astra’s performance at about one-fifth Astra’s task cost. On AutomationBench 1.0.6, which tests workflows involving 47 tools, Sol scored 2.2 percentage points above Opus 5.5 and 4.8 points above GPT-6 Sol at medium reasoning effort.

Computer use, science, and factuality On the OSWorld 2.0 offline computer-use test, Sol scored seven percentage points above GPT-6 Sol at maximum reasoning effort. It finished within 2.1 points of Astra at roughly one-seventh of Astra’s cost per task.

In Terminal-Bench Science 0.1, OpenAI reported an average cost of $5.47 per task for Sol at maximum effort, compared with $23.21 for Opus 5.5 and $23.80 for Astra. Astra remained the highest-scoring model in that test, at 68.1%.

OpenAI also reported that, at low reasoning effort, the share of tested answers containing a factual error fell from 11.4% with GPT-6 Sol to 7.7% with GPT-6.1 Sol. In a separate broken-search-tool test at maximum effort, Sol failed to disclose the problem in 2.1% of cases, versus 4.9% for GPT-6 Sol and 28.7% for GPT-6 Luna. OpenAI says these deliberately difficult tests do not represent typical usage.

Availability Developers can use the model through the API as gpt-6.1-sol. As of its September 29 launch, it is also available in ChatGPT Work and Codex for Plus, Pro, Business, Enterprise, and Edu users. OpenAI says GPT-6.1 Sol is not yet available in Chat.