GPT-6 Luna

GPT-6 Luna is OpenAI's most efficient GPT-6 model, released September 22, 2026 next to GPT-6 Sol and positioned below it in the family. The API id is gpt-6-luna. OpenAI cut the price in half against GPT-5.6 Luna's promotional rate: $0.10 per 1M input tokens and $0.50 per 1M output tokens, with cached input at $0.01. It keeps the GPT-6 context window at 1,050,000 tokens and supports reasoning effort from none through max.

The previous cheap tier is GPT-5.6 Luna, and the mid-tier sibling is GPT-5.6 Terra.

Key Features

• Cheapest GPT-6 row: At $0.10/$0.50 with $0.01 cached input, Luna is the tier OpenAI positions for bulk classification, routing, and long agent loops.
• Full GPT-6 context: 1,050,000-token context with 922,000 max input and 128,000 max output, the same envelope as Sol and Astra.
• Effort control, including none: none, low, medium (default), high, xhigh, and max. If you want function calling on Chat Completions, effort has to stay at none.
• Free-tier access: OpenAI says Free and Go users can reach GPT-6 Luna in the desktop app, while paid plans get Sol and Luna in ChatGPT Work and Codex.
• Already in Copilot: GitHub's supported-models table lists GPT-6 Luna as GA next to GPT-6 Sol and GPT-6 Astra.

Model Specifications

Pricing

From OpenAI's GPT-6 Luna model card, Standard text tokens:

The same 272K-token surcharge applies: prompts above that size are billed at 2x input and cache rates and 1.5x output for the full request. Batch and Flex are 50% of Standard, and Fast mode is 2x.

Use Cases

• High-volume classification and routing: OpenAI places Luna under the fast tier, and the $0.10 input rate makes fan-out patterns affordable.
• Cost-bounded coding: OpenAI reports DeepSWE v1.1 at 66.6% for GPT-6 Luna at max effort, roughly level with Claude Opus 5 and Fable 5 at medium effort, at about 93% less cost per task than Opus 5.
• Business workflows on a budget: On AutomationBench, OpenAI says GPT-6 Luna at high effort improves on GPT-5.6 Luna by 5.4 percentage points at 58% lower cost per task.
• Computer use at the low end: OpenAI reports GPT-6 Luna at max effort exceeding GPT-5.6 Sol at medium effort on OSWorld 2.0 offline, at about one tenth of the cost.

Limitation: the efficiency shows up in cost per task, not in raw scores. Artificial Analysis placed GPT-6 Luna at a 37.3 intelligence index, below Sol, Astra, and the current Claude frontier. On tasks that need deep reasoning, the savings can vanish when you burn more tokens to get a correct answer.

Getting Started

1. Set model to gpt-6-luna on the Responses API.
2. Try medium effort first, which is the default; drop to low or none for pure extraction jobs.
3. Cache shared prefixes. At $0.01 per 1M cached tokens, reuse is 10x cheaper than fresh input.
4. Keep long prompts under 272K input tokens unless you accept the multiplier.

Frequently Asked Questions

What is the difference between GPT-6 Luna and GPT-6 Sol?

Both share the 1.05M context and the none to max effort scale. Luna is 20x cheaper on input and built for volume and latency; Sol is built for difficult coding and agentic work. OpenAI prices them at $0.10/$0.50 and $2/$10 respectively.

Is it free to use?

Not as a product, but OpenAI says Free and Go users can access GPT-6 Luna in the ChatGPT desktop app, which makes it the first GPT-6 model broadly reachable without a paid plan.

Should I migrate from GPT-5.6 Luna?

If your workload is cost-sensitive and mostly high-volume, yes: the list price is 50% lower for the same context window. Re-test accuracy on your own evals first.

Alternatives

• GPT-6 Sol: the step up when Luna's scores are not enough.
• GPT-6 Astra: OpenAI's flagship for the hardest tasks.
• GPT-5.6 Luna: the previous cheap tier.

Tips

1. Route by difficulty. Send easy traffic to none or low effort and reserve xhigh and max for the small share of hard requests.
2. Watch the cache-write price. At $0.125 per 1M tokens it is cheap, but it is still 25% above a plain input token, so cache prefixes you will actually reuse.
3. Do not assume a cheaper model is a worse one. On several OpenAI benchmarks Luna at max effort beats the previous generation's Sol at medium effort.

Conclusion

GPT-6 Luna is OpenAI's volume model: a 1.05M context window, effort control up to max, and a $0.10/$0.50 price that undercuts everything else in the GPT-6 line. Read the announcement and the API model card, then route only the traffic that genuinely needs more than this tier to GPT-6 Sol.