Coding, document analysis, computer use, and multi-step agents in one model
GPT-6.1 Sol is OpenAI's Sol model for complex coding, computer use, and professional work. It supports agents that inspect code, use tools, read detailed documents, and carry a task through multiple steps.
Its 1,050,000-token context window and adjustable reasoning effort help developers work across large codebases and long source material. OpenAI reports gains over GPT-6 Sol on software engineering, document understanding, business automation, and computer-use evaluations.
Capabilities for development teams and agents working across code, documents, and applications.
01
GPT-6.1 Sol can investigate a codebase, trace failures, edit multiple files, and verify changes across a longer development task. In OpenAI's DeepSWE v1.1 evaluation, it reached GPT-6 Astra's result and improved on GPT-6 Sol's best score by 6.4 percentage points at a lower reasoning setting.
02
GPT-6.1 Sol works with detailed PDFs containing tables, charts, diagrams, and fine print. It also coordinates multi-step business workflows that use many tools. OpenAI reports stronger results than GPT-6 Sol on GDP.pdf and a 4.8-point improvement on AutomationBench at medium reasoning effort.
03
GPT-6.1 Sol can interpret visual context and work through application interfaces as part of an agent workflow. On the offline OSWorld 2.0 evaluation, OpenAI reports a seven-point gain over GPT-6 Sol at maximum reasoning effort. Teams can use this capability for tasks that combine screenshots, interface actions, and verification.
04
GPT-6.1 Sol accepts up to 1,050,000 context tokens and produces up to 128,000 output tokens. Developers can choose low, medium, high, xhigh, or max reasoning effort to match the depth of work. OpenAI's factuality evaluation also shows more reliable answers at low reasoning effort. This range supports large repositories, lengthy reports, and extended agent sessions.
Use GPT-6.1 Sol to explore a repository, implement features, resolve bugs, and review changes across multi-file development tasks.
Build agents that plan steps, call tools, inspect results, and continue work with the context of earlier actions.
Analyze detailed PDFs, combine evidence from charts and text, and prepare summaries grounded in source material.
Use screenshots and application state to examine interfaces, carry out workflows, and check observable results.
Coordinate repeatable business processes across tools, documents, and handoffs while tracking each step.
Work through terminal-based analysis, simulation, and research tasks that require code and iterative checks.
| Capability | GPT-6.1 Sol | GPT-6 Sol | GPT-6 Astra |
|---|---|---|---|
| Model focus | Complex coding, computer use, and professional workflows | Complex coding and agent workflows | The most demanding end-to-end work |
| DeepSWE v1.1 | Matches Astra; 6.4 percentage points above Sol's best score | 6.1 Sol exceeds its best score by 6.4 percentage points | Matches 6.1 Sol |
| OSWorld 2.0 offline (max effort) | 7 percentage points above Sol; within 2.1 points of Astra | 6.1 Sol scores 7 percentage points higher | 2.1 percentage points above 6.1 Sol |
| Terminal-Bench Science 0.1 (max effort) | More than 2× Sol's score | 6.1 Sol scores more than 2× this result | Top reported score: 68.1% |
| Factual error rate (low effort) | 7.7% | 11.4% | 6.1 Sol is within 1.9 percentage points across tested efforts |
| Official price · Standard input / 1M tokens | $2.00 | $2.00 | $10.00 |
| Official price · Cached input / 1M tokens | $0.10 | $0.20 | $1.00 |
| Official price · Cache writes / 1M tokens | $2.50 | $2.50 | $12.50 |
| Official price · Standard output / 1M tokens | $10.00 | $10.00 | $50.00 |
Find GPT-6.1 Sol alongside other PoYo models and review its coding, document, computer-use, and agent capabilities in one place.
Use OpenAI's published benchmarks to compare GPT-6.1 Sol with GPT-6 Sol and GPT-6 Astra for the tasks your team handles.
Map long-context inputs, reasoning settings, and tool steps to your coding and agent workflows before integration.
Follow GPT-6.1 Sol's coming-soon status on PoYo and use this page to prepare for its availability.
