GPT-6 Luna
OpenAI·
LLMssmallcloud
Overview
Speed- and volume-tuned tier of the GPT-6 family, launched together with GPT-6 Sol. $0.10 / $0.50 per million input/output tokens, a tenth of GPT-5.6 Luna's price, with cached reads at 10% of input. 1M-token context, text and image input. Available in the API, Codex and ChatGPT Work; Free and Go users get it in the desktop app.
Capabilities and innovations
Agentic codingComputer useTool use / function callingReasoning effort control1M token context windowAPI: gpt-6-lunaSol / Luna GPT-6 tiers
Benchmarks
| Benchmark | Score | Source |
|---|---|---|
| Terminal-Bench 2.x | 73.03% | independent Terminal-Bench 2.1 73.03%, measured by Vals AI at max effort (accessed 2026-09-27). Same Vals AI harness as the GPT-5.6 Luna entry (79.03), so the drop is like-for-like. |
| DeepSWE v1.1 | 66.6% | vendor DeepSWE v1.1 66.6 at max effort, OpenAI launch post. Vendor self-report over its own model. |
| ARC-AGI-1 | 86.7% | independent ARC-AGI-1 86.7%, verified by ARC Prize at max effort (results dated 2026-09-22). |
| ARC-AGI-2 | 59.3% | independent ARC-AGI-2 59.3%, verified by ARC Prize at max effort (results dated 2026-09-22). ARC-AGI-3: 0.1% (standard harness, max). |
- GPQA Diamond: independent evaluation pending — Not reported by OpenAI; Artificial Analysis shows only the composite index, no per-benchmark value (checked 2026-09-27).
- Humanity’s Last Exam (no tools): independent evaluation pending — Not reported by OpenAI; no per-benchmark value on Artificial Analysis yet (checked 2026-09-27).
- SWE-bench Pro: not reported by the vendor — OpenAI reports DeepSWE v1.1 instead, recorded in deepswe_v1_1.
- MMLU-Pro: not reported by the vendor
- Terminal-Bench 4.0: independent evaluation pending — Not on the official tbench.ai 4.0 board and no per-benchmark value on Artificial Analysis yet (checked 2026-09-27).
API pricing
$0.1 per million input tokens, $0.5 per million output tokens (USD, provider list price, checked 2026-09-27)
Links
More from OpenAI
Data curated by AI Model Timeline. See the methodology for admission criteria, benchmark eras and source priority.