WordOnThisWordOnThisSearch
Submit
GPT-6.2 Sol logo

GPT-6.2 Sol

Runs everyday software engineering and tool-using agents at scale.

Visit website
GPT-6.2 Sol official website preview
Loading website screenshot…The official website preview is loading.

What is GPT-6.2 Sol?

OpenAI designs GPT-6.2 Sol for everyday software engineering and agent workflows running at scale. Charged per token, it processes text and images across a 1,050,000-token context window (up to 922,000 input and 128,000 output tokens) and emits text or structured outputs. Through the Responses API, it works with computer use, web search, file search, and an optional multi-agent beta. Developers can apply it to implementing and testing code changes as well as running tool-driven automation.

What can GPT-6.2 Sol do?

Large context with structured output

Handles up to 922,000 input tokens and 128,000 output tokens across text and image inputs, billed per token.

Computer and web tool integration

Connects through the Responses API to computer use, web search, and file search tools for running agent workflows.

Optional Responses multi-agent beta

Supports coordinating multiple agents within the Responses API for parallel or staged engineering tasks.

Who is GPT-6.2 Sol for?

  • Day-to-day feature and bug implementationTeams can assign it to write application changes, update tests, and iterate on repository tasks.
  • Scaled tool-using agentsOrganizations can run agents that combine web search, file lookup, and computer actions.
  • Automated pull-request preparationDevelopers can test it on drafting patches and verification scripts across active branches.

How much is GPT-6.2 Sol?

View official pricing ↗
paid

OpenAI standard API, up to 272,000 input tokens: USD $2 per million uncached input tokens and $10 per million output tokens. Above 272,000 input tokens: USD $4 per million uncached input tokens and $15 per million output tokens for the entire request.

Frequently asked questions

What is the OpenAI API model ID?

Use gpt-6.2-sol to call this model.

How did it perform on Terminal-Bench 4.0?

Vals AI reports 60.61% on Terminal-Bench 4.0 with the Mini-SWE-agent harness. The score applies to that evaluated setup. Source: https://www.vals.ai/benchmarks/terminal-bench-4

What people are saying

Public experiences and discussions, with links to the originals.

No discussions collected yet.

Explore 11 products that share at least one second-level category.

View all 11 alternatives

Help keep this listing accurate

Found something incorrect? Suggest a correction for our editors. It’s free; no account or email is needed.