WordOnThisWordOnThisSearch
Submit
GPT-6 Luna logo

GPT-6 Luna

Delivers cost-efficient structured outputs for high-volume routing and code tasks.

Visit website
GPT-6 Luna official website preview
Loading website screenshot…The official website preview is loading.

What is GPT-6 Luna?

OpenAI built GPT-6 Luna as a cost-efficient model for focused, high-volume workloads and also benchmarks it for software engineering. Billed per token, it takes text and images across a 1,050,000-token context window (up to 922,000 input and 128,000 output tokens) and returns text or programmable structured outputs. It works with Responses API tools such as web search, file search, and computer use, as well as the Batch API. Examples include classification, data extraction, routing, bounded repository tasks, and routine code transformations.

What can GPT-6 Luna do?

Programmable structured outputs

Formats responses to match schemas for parsing in high-volume pipelines, charged per token.

1,050,000-token context and Batch API

Supports up to 922,000 input and 128,000 output tokens from text and image inputs, plus asynchronous Batch API runs.

Responses API tool support

Works with web search, file search, and computer use in the Responses API when focused tasks need external lookups.

Who is GPT-6 Luna for?

  • Classification, extraction, and routingPipelines can classify incoming tickets, extract fields into JSON, and route requests at volume.
  • Bounded repo tasks and code transformsTeams can run routine migrations, lint fixes, or scoped edits across files.
  • Asynchronous Batch API processingDevelopers can queue large sets of document extraction or tagging jobs through the Batch API.

How much is GPT-6 Luna?

View official pricing ↗
paid

OpenAI standard API, up to 272,000 input tokens: USD $0.10 per million uncached input tokens and $0.50 per million output tokens. Above 272,000 input tokens: USD $0.20 per million uncached input tokens and $0.75 per million output tokens for the entire request.

Frequently asked questions

What is the OpenAI API model ID?

Use gpt-6-luna to call this model.

How did it perform on Terminal-Bench 4.0?

Vals AI reports 13.64% 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.