Model release

GPT 6 Luna Decisions API General Availability

OpenAI announced GPT 6 Luna general availability on Oct 22, 2026. The Decisions API now supports 400 questions per request with image and JSON…

Released Published 3 min readModel evidence ↗
GPT 6 Luna release illustration — OpenAI

General Availability Announcement

The official changelog records the general availability of the Decisions API with the gpt-6-luna model on October 22, 2026. This date marks the transition from public beta to general availability. The source explicitly states that each request now supports up to 400 questions, an increase from the 200-question limit during the public beta phase.

Documented Feature Changes

The release notes list three specific functional additions for the Decisions API. First, users can cache shared input across requests to receive prompt-cache discounts. Second, images can be included in individual question instructions. Third, structured data can be passed directly using the inputjson parameter. These features are documented as part of the current API version v1/decisions.

  • Prompt-cache discounts apply to shared inputs across requests.
  • Images are supported within individual question instructions.
  • Structured data is accepted via the inputjson field.

Evaluation Procedure and Limits

Engineers should evaluate the model by constructing test cases that utilize the new 400-question limit. Measure latency and accuracy when passing structured data through inputjson compared to previous methods. Validate image processing by including visual assets in question instructions and comparing output consistency. The source confirms these capabilities exist but does not provide benchmark scores or guaranteed performance improvements over previous iterations. It remains unknown how these changes affect total cost beyond the stated discount mechanism for cached inputs.

Artificial Analysis benchmark review

Benchmark review updated on . These scores apply to the linked artifact and recorded methodology; they do not establish the performance of a different runtime or quantized model.

Artificial Analysis source measurements for the linked model artifact.
MeasurementRecorded resultMeasured
Artificial Analysis Intelligence Index (v4.3) ↗38.1 index_points2026-10-01

Review the full benchmark provenance and source methodology →

Primary sources

QUICK ANSWERS

Frequently asked questions

What is the maximum number of questions per request?

The source states that each request supports up to 400 questions, which is an increase from the 200-question limit in the public beta.

Does the source confirm local installation requirements?

No. The source only records that open weights are not confirmed. It does not establish whether local installation is possible or if API access is mandatory.

More from the journal

Grok Voice Transcribe 2.0 Release Details →

In September 2026, xAI announced the release of Grok Voice Transcribe 2.0. The official changelog identifies grok-voice-transcribe-2.0 as the new default for speech-to-text tasks. Developers can still select grok-voice-transcribe-1.0. This article outlines the confirmed specifications and provides a method to test performance differences between versions.