OpenAI: GPT-6.1 Sol (batch)

GPT-6.1 Sol is a proprietary reasoning model in OpenAI's GPT-6 series, released in September 2026 as a cost-efficient upgrade to GPT-6 Sol. It targets agentic coding, computer use, document-heavy professional work, and multi-step workflow automation, with OpenAI positioning it close to GPT-6 Astra on demanding tasks while standard API input and output list at one-fifth of Astra's rates (batch and flex tiers are half of standard). The model is served only through OpenAI's platform under the ID gpt-6.1-sol, including the Batch API variant for asynchronous workloads at reduced per-token pricing.

Architecturally it is a closed-weight transformer with configurable reasoning effort (low through max; medium is the default), a 1.05 million-token context window, up to 128K completion tokens, and an April 2026 knowledge cutoff. It accepts text and image inputs and returns text, with tool use, structured outputs, computer-use, and web-search integrations supported on the Responses API. Parameter count and training compute are not disclosed; training follows the same data and methods described for GPT-6 Astra in OpenAI's system-card addendum.

Reported evaluations emphasize software engineering and agents: DeepSWE v1.1 at 75.2% (high reasoning, with tools), AutomationBench at 31.7% (medium), and strong multimodal and exam scores such as MMMU-Pro and LiveBench at max effort. Independent Artificial Analysis runs at max effort report 52.92% on Humanity's Last Exam and 95.4% on GPQA Diamond (Epoch AI Benchmarking Hub). The model ships in ChatGPT Work and Codex for eligible paid plans and is not offered as a general consumer Chat model or for local weight download.