GPT-6.1 Sol cuts the price of near-Astra agentic coding to a fifth

OpenAI lists GPT-6.1 Sol at $2 per million input tokens, $10 output and $0.10 cached input. GPT-6 Astra is listed at $10, $50 and $1 cached. On DeepSWE v1.1, which OpenAI describes as complex software-engineering tasks in real codebases, it says Sol matches 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. On AutomationBench 1.0.6, OpenAI reports Sol 2.2 points above Opus 5.5 at medium reasoning effort at roughly a third of the cost. On OSWorld 2.0's offline set it reports Sol within 2.1 points of Astra at maximum effort at roughly one-seventh the cost per task.

One safety figure is worth reading in the harness context. On a test of whether an agent tells the user its search tool is broken, OpenAI reports Sol fails to disclose the problem in 2.1% of cases, against 4.9% for GPT-6 Sol and 1.5% for Astra. OpenAI says the tasks are selected to elicit failures and do not represent typical use.

Sol is available in ChatGPT Work and Codex to Plus, Pro, Business, Enterprise and Edu users, and in the API as gpt-6.1-sol. It is not yet in Chat. A faster Ultrafast variant is promised "in the coming days".

Why it matters: The cached-input price is the number for long-running agent loops that resend the same context, and it is 95% below Sol's standard input rate. If the DeepSWE claim holds on your own codebase, it changes which tier you route routine implementation work to. These are vendor-run benchmarks, so test it on your own tasks before moving a workload.