GPT-6 Luna
OpenAI's most efficient model, built for focused, high-volume tasks. It is one of the cheapest models in this list to run, but it trails the larger GPT-6 models by a wide margin on hard and agentic tasks.
OpenAILanguage modelReleased September 2026Data checked October 18, 2026
This model is not in the current people's rankings: it is either new and still collecting votes, or ranked below the top 12. See the AI leaderboard for the models people rate highest.
On paper
Fixed-test rankings from the benchmarks tab: Speed #8, Reasoning #21.
Strengths and weaknesses
- Lowest Artificial Analysis cost per task of any paid model here ($0.07)
- Low price ($0.10 / $0.50 per million tokens) with a 1.05M-token context window
- Scores 59.3% on ARC-AGI-2 at about 6 cents per task
- Weak on agentic terminal work (16.4% on Terminal-Bench 4.0 in Codex)
- Low LMArena text and writing scores compared with the bigger GPT-6 models
Best for
- Cheap high-volume tasks
- Simple chat and data extraction
- Sorting and tagging content
Every score we have
Each bar compares this model with the best result any model on this page has on the same test.
- LMArena text arena (human votes)1,442.8Arena score 1442.8 (95% CI 1438.2 to 1447.4); 49,617 votes; entry "gpt-6-luna-max"; position 91 of 416 on the full board
- LMArena creative writing1,410Arena score 1410.0 (95% CI 1402.9 to 1417.2); 11,078 votes; entry "gpt-6-luna-max"; position 91 of 414 on the full board
- LMArena WebDev (coding)1,578.2Arena score 1578.2 (95% CI 1572.0 to 1584.3); 20,045 votes; entry "GPT-6 Luna (Max)"; position 35 of 144 on the full board
- LMArena vision1,241Arena score 1241.0 (95% CI 1228.8 to 1253.1); 2,833 votes; entry "gpt-6-luna-max"; position 65 of 165 on the full board
- LMArena Agent Arena0.030IPS score 0.0301 (95% CI 0.0255 to 0.0347); 163,815 sessions; entry "GPT-6 Luna (Max)"; position 18 of 52 on the full board
- Artificial Analysis Intelligence Index38.138.1 on the Artificial Analysis Intelligence Index v4.3.2 (AA shows 38), setting "GPT-6 Luna (max)"
- Humanity's Last Exam (Artificial Analysis)38.5%38.5% on Humanity's Last Exam (AA, 2,158 text-only questions), setting "GPT-6 Luna (max)"
- Terminal-Bench 4.0 (Artificial Analysis run)12.6%12.6% on Terminal-Bench 4.0 as run by AA (mini-swe-agent harness), setting "GPT-6 Luna (max)"
- Long-context reasoning (AA-LCR)83.3%83.3% on AA-LCR (long-context reasoning), setting "GPT-6 Luna (max)"
- MMMU-Pro, image understanding79.7%79.7% on MMMU-Pro (image understanding, AA), setting "GPT-6 Luna (max)"
- AA-Omniscience, accuracy minus hallucination0.70.7 on the AA-Omniscience Index (knowledge accuracy minus hallucination, range -100 to 100), setting "GPT-6 Luna (max)"
- Output speed, tokens per second120 t/s120 output tokens per second (AA median), row "GPT-6 Luna (max)"
- Cost per Artificial Analysis test task$0.07$0.07 to run one AA Intelligence Index task, row "GPT-6 Luna (max)"
- Artificial Analysis Coding Agent Index41.141.1 on the AA Coding Agent Index v1.5 (DeepSWE v1.1, Terminal-Bench 4.0, SWE-Atlas-QnA), best listed setup "Codex - GPT-6 Luna (max)"
- Terminal-Bench 4.0 leaderboard16.4%16.4% resolved on Terminal-Bench 4.0 (plus or minus 2.7, 95% CI) with Codex at max effort
- ARC-AGI-2 puzzles59.3%59.31% on ARC-AGI-2 (semi-private set), best listed setting "GPT-6 Luna (Max)", $0.06 per task
- ARC-AGI-3 puzzles0.2%0.19% on ARC-AGI-3 (semi-private, standard harness; "provider adapter" runs excluded), best listed setting "GPT-6 Luna (Medium)"
Compare with
- Claude Opus 5.5Anthropic's flagship for long-running agentic coding and knowledge work, and the model Anthropic suggests starting with for most jobs.
- Claude Sonnet 5.5Anthropic's mid-priced model, which Anthropic describes as its best mix of speed and intelligence.
- Claude Fable 5.1Anthropic's top tier for the hardest reasoning and long-horizon agent work, which Anthropic recommends when Opus 5.5 still falls short.
- Claude Haiku 5.5Anthropic's smallest and fastest current model, built for high-volume, latency-sensitive jobs like classification, extraction and routing.