OpenAI GPT-6.1 Sol (Vision Reasoning)

Overview

GPT-6.1 Sol is a vision-capable reasoning model from OpenAI. It combines image understanding with long-context text reasoning, then returns a structured text answer you can act on. You can also let it request tool calls when a task needs search, computation, or multi-step work. It’s useful when you need fewer handoffs between “see,” “think,” and “do” in one run.

What you can build

• Screenshot and UI troubleshooting that reads what’s on screen and explains next steps
• Image-grounded coding help, like debugging from screenshots of logs or dashboards
• Visual document Q&A for charts, tables, diagrams, and fine print
• Agent-style workflows that decide when to call tools and when to answer directly
• Multi-image comparisons, like “spot the differences” or change tracking across versions

Inputs

• A required written instruction in plain text that tells the model what to do.
• Optional image files that provide visual context for the instruction. You can attach multiple images in one run, up to 10 total.
• Optional user and session identifiers (string or number) to keep chat history grouped the way you expect.
• Optional system-level instructions that set global rules for the run. Keep them under 5,000 characters.
• Optional reasoning effort selection (low, medium, high, xhigh, or max) to trade off cost against depth.
• Optional verbosity selection (low, medium, or high) to control how much detail the model writes.
• Optional maximum output length cap to limit how many tokens the model can generate, up to 128,000 tokens.
• Optional structured conversation messages as an alternative to a single instruction. Use either a single instruction or message history, not both.
• Optional tool definitions if you want the model to call functions you provide. You can also control whether tool calls are allowed, required, or disabled.
• Optional continuation data for multi-step runs. Provide the prior task token, then include tool results for the tool calls from that prior step.

Outputs

The model returns a text response that follows your instruction and references the provided images when relevant. If tool use is enabled, it can return one or more tool call requests instead of, or before, a final answer. When you continue a prior run, the output includes the next assistant turn after it processes your provided tool results.

Limitations

• GPT-6.1 Sol does not support “none” or “minimal” reasoning. Use low effort when you need lighter reasoning.
• The model has a knowledge cutoff of April 30, 2026. It may be wrong about newer events.
• Long inputs can still fail if you exceed the model’s 1,050,000-token context window.
• Image understanding depends on input quality. Blurry photos, heavy compression, glare, tiny text, or scanned pages can cause mistakes.
• Tool use depends on what tools you provide and how you configure them. If tools are missing or misconfigured, it may not complete the workflow.
• Fine-tuning is not supported for this model.

Safety & compliance

OpenAI applies the same safeguards stack to GPT-6.1 Sol as GPT-6 Astra, with heightened focus on high-risk domains like cybersecurity. The model may refuse or constrain requests that enable wrongdoing, including instructions for malware, exploitation, credential theft, or other harmful activity. You’re responsible for ensuring your prompts, images, and tool integrations follow OpenAI’s usage policies and any applicable laws, especially for regulated content and sensitive personal data.