GPT-5.6 Luna

// Overview

GPT-5.6 Luna is the cost-optimized model in OpenAI's GPT-5.6 series, first shown as a limited preview on June 26, 2026 and publicly released on July 9, 2026 alongside the flagship Sol and the mid-tier Terra. OpenAI positions Luna as its fast, affordable option in the GPT-5.6 family, aimed at high-volume background tasks, batch processing, embedded applications, classification, and simple agent steps.

Luna accepts text and image input, supports a 1,050,000-token context window with up to 128,000 output tokens, and carries a February 16, 2026 knowledge cutoff. On July 30, 2026 OpenAI cut Luna's headline pricing by 80%, from $1.00 / $6.00 to $0.20 per 1M input tokens and $1.20 per 1M output tokens, citing efficiency gains from GPT-5.6's own use in optimizing OpenAI's production stack.

In the GPT-5.6 naming system, the number identifies the generation and the tier names — Sol, Terra, Luna — identify durable capability tiers that can advance on their own cadence.

On August 6, 2026 OpenAI shipped an updated Luna and made it the everyday default model in ChatGPT for Free and Go users, replacing GPT-5.5 Instant. OpenAI's system card distinguishes this snapshot as the 'August' version; the July version remains in use in Codex and ChatGPT Work. The August release cuts factual error rates by over 60% on high-stakes medical, legal, and financial prompts and by roughly 30% on OpenAI's other two hallucination prompt sets, and improves on every HealthBench evaluation over GPT-5.5 Instant despite Luna's smaller size. Under OpenAI's Preparedness Framework the August release is treated as High capability in both Cybersecurity and Biological and Chemical domains.

// Benchmarks

This model's scores

1. Terminal-Bench 2.182.5%
2. HealthBench (length-adjusted, August release)53.3
3. HealthBench Hard (length-adjusted, August release)28.7
4. HealthBench Consensus (length-adjusted, August release)94.8
5. HealthBench Professional (length-adjusted, August release)44.1

Scores on a 0–100 scale (25-point gridlines); higher is better. Each benchmark links to its published source.

// Pricing

Reduced from $1.00 / $6.00 on July 30, 2026 (80% price cut). OpenAI attributes the reduction to efficiency gains from using GPT-5.6 internally to rewrite production code.

Pricing source ↗

// Strengths

• OpenAI's lowest headline pricing in the GPT-5.6 family — $0.20 input / $1.20 output per 1M tokens after the July 30, 2026 cut
• Reasoning-model architecture with text and image input
• Designed for high-volume background, batch, and embedded workloads
• Ships with the GPT-5.6 series safety stack for production deployments
• Serves as ChatGPT's everyday default for Free and Go users from August 6, 2026
• August 2026 release cuts factual error rates by over 60% on high-stakes prompts versus GPT-5.5 Instant

// Best for

• Reach for it for high-volume batch processing, classification, and embedded application work.
• Reach for it for simple agent steps where cost per call matters more than reasoning depth.
• Reach for it when a workload's economics require the lowest GPT-5.6 per-token rate.

// How to access

// GPT (Flagship / Thinking) — every version

The full lineage of the GPT (Flagship / Thinking) line, newest first. Every version has its own page — click any to compare specs, benchmarks and pricing.