GPT 6 Luna

What is GPT 6 Luna?#

GPT 6 Luna is the small, fast model in OpenAI's GPT-6 family, released on September 22, 2026 alongside GPT 6 Sol. OpenAI calls it its fastest and most cost-efficient model yet. It runs on Unifically as openai/gpt-6-luna. It takes text and image input and returns text, with a 1,050,000-token context window, a 128,000-token max output, and knowledge up to May 18, 2026. OpenAI's API price is half of GPT 5.6 Luna's, and it keeps the same reasoning effort range, from none up to max. It is built for the calls you make thousands of times a day, and at max effort it still gets close to Claude Opus 5 on agentic coding.

Key features of GPT 6 Luna#

Near-Opus coding at a fraction of the cost

On DeepSWE v1.1, Luna scores 66.6% at max effort for $0.22 a task. Claude Opus 5 at medium effort scores 68.9% for $3.29 and Claude Fable 5 65.4% for $6.09, so Luna costs 93% less per task than Opus 5 and 96% less than Fable 5. GPT 5.6 Luna scores 62.2%.

GPT 5.6 Luna's agent score for 94% less

On Agents' Last Exam, which runs long professional workflows across 55 fields, Luna scores 50.9% at max effort for $0.15 a task. GPT 5.6 Luna scores 50.4% on the same test and spends $2.57 a task to get there.

Fewer factual errors than GPT 5.6 Sol

On OpenAI's set of difficult factual prompts, Luna at max effort answers with a factual error 7.6% of the time, below GPT 5.6 Sol at 8.5% and GPT 5.6 Luna at 12.0%, for about $0.012 a prompt against $0.87 for GPT 5.6 Sol.

Built for throughput

Artificial Analysis measures about 132 output tokens per second at max effort and 143 at medium, and ranks Luna 6th of 174 models in its class on the Intelligence Index with a score of 37. A full run of that index costs about $0.07 per task, against $0.18 for GPT 5.6 Luna.

Best for#

Use cases#

Route the bulk of your traffic through Luna: ticket tagging, entity extraction, document summaries, and short assistant replies. Use it as the worker model inside an agent system, handling the many small tool calls while GPT 6 Sol plans and checks the result. Read screenshots, receipts, and scanned forms in bulk. For coding, use it for draft patches, quick explanations, and test scaffolding, where a near-Opus score at a few cents a task changes what is affordable to automate. Batch jobs that share one long prompt template get the most from caching, since cached reads bill at a tenth of the input rate.

Limitations#

Luna is the small model and it shows on harder work. Artificial Analysis measures its Coding Agent Index at 41, two points below GPT 5.6 Luna, with lower SWE-Atlas scores, and its knowledge-work evals dropped about 75 Elo on GDPval and 45 on AA-Briefcase. It also spends more tokens per task than GPT 5.6 Luna, about 51,000 against 41,000 on that index. On the Intelligence Index it sits below Gemini 3.8 Flash and DeepSeek V4.1 Flash.

The strong benchmark numbers come from max effort, which means long thinking before the first answer token: Artificial Analysis measures about 109 seconds on its hard prompts. For latency-sensitive calls use low or medium effort and expect lower scores.

OpenAI rates Luna High for cybersecurity and biology risk, so it runs behind the same safeguards as GPT 5.6. OpenAI also says some of its jailbreak robustness reflects a broader tendency to refuse, including some legitimate requests.

Long prompts cost more. Once a request's prompt passes 272,000 tokens, the whole request bills at the long-context rate: double on input, cache read, and cache write, and 1.5x on output.

When to use GPT 6 Luna#

Use Luna when volume and cost are the constraint: the calls you make thousands of times a day, the small agent steps, and the batch jobs where price per request decides whether the job is worth running. When quality on hard reasoning or coding starts to bind, step up to GPT 6 Sol, and keep GPT 6 Astra for the hardest tasks.

API examples#

Call GPT 6 Luna from any language by POSTing to /v1/chat/completions, the OpenAI-compatible endpoint shared by every language model on the platform. Full parameter docs live at docs.unifically.com/models/llm/openai/gpt-6-luna.

curl -X POST https://api.unifically.com/v1/chat/completions \
  -H "Content-Type: application/json" \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -d '{
    "model": "openai/gpt-6-luna",
    "messages": [
      { "role": "user", "content": "Classify this support ticket as billing, bug, or feature request: My card was charged twice." }
    ]
  }'

The response comes back synchronously with the completion. Set "stream": true to receive tokens as they generate, and "reasoning_effort": "low" for the fastest replies.