Sign Up

GPT-5.6 Luna

OpenAI
Announcement
GPT-5.6 Luna launch post cover

Cost/1M tokens

input
$0.20
cached
$0.02
output
$1.20
cache write
$0.25
Context
1.1Mtokens
Released
4mo ago

Benchmarks

DeepSWE is the primary coding benchmark where an exact model evaluation exists. Arena leaderboards and other sourced evaluations appear below; missing scores are not zero.

How it ranks

Against its predecessor, similarly priced models and the models ranked around it.

DeepSWE 1.1

Best pass@1 per model · Higher is better

#10 of 28

  1. 74.1%
    GPT-6 Astra
  2. 73.8%
    Gemini 3.8 Flash
  3. 73.6%
    Claude Opus 5
  4. 72.7%
    GPT-5.6 Sol
  5. 69.9%
    Claude Fable 5
  6. 69.6%
    GPT-5.6 Terra
  7. 69.0%
    GLM 5.3
  8. 68.5%
    Kimi K3
  9. 67.5%
    Grok 4.6
  10. 67.2%
    GPT-5.6 Luna

Arena Text

Best Arena rating per model · Higher is better

Arena #77

  1. 1481
    GPT-5.5Predecessor
  2. 1455
    Gemini 3.5 Flash LiteNearby rank
  3. 1454
    Hy3Nearby rank
  4. 1453
    GPT-5.6 Luna
  5. 1452
    Gemma 4 31BNearby rank
  6. 1450
    Claude Opus 4.1Nearby rank
  7. 1435
    Beam 1.0 501B A23BSimilar price
  8. 1416
    DeepSeek V3.1 TerminusSimilar price
  9. 1391
    MiniMax M2.5Similar price
  10. 1378
    GLM 4.6VSimilar price

-28 vs GPT-5.5 (predecessor) · Ahead of 4 of 4 similarly priced models

Arena WebDev

Best Arena rating per model · Higher is better

Arena #62

  1. 1622
    GLM 5.3Similar price
  2. 1621
    DeepSeek V4.1 FlashSimilar price
  3. 1522
    Claude Sonnet 4.6Nearby rank
  4. 1520
    GPT-5.6 TerraNearby rank
  5. 1519
    GPT-5.6 Luna
  6. 1515
    Qwen3.7 MaxNearby rank
  7. 1512
    GPT-5.5Predecessor
  8. 1437
    Beam 1.0 501B A23BSimilar price
  9. 1386
    MiniMax M2.5Similar price

Level with GPT-5.5 (predecessor, overlapping 95% CIs) · Ahead of 2 of 4 similarly priced models

Arena Vision

Best Arena rating per model · Higher is better

Arena #50

  1. 1285
    GPT-5.5Predecessor
  2. 1263
    DeepSeek V4.1 FlashSimilar price
  3. 1258
    GPT-6 SolNearby rank
  4. 1256
    Seed 2.0 ProNearby rank
  5. 1255
    GPT-5.6 Luna
  6. 1252
    Kimi K2.5Nearby rank
  7. 1251
    GPT-5.1Nearby rank
  8. 1161
    GLM 4.6VSimilar price
  9. 1145
    Llama 4 MaverickSimilar price
  10. 1127
    Mistral Small 3.1 24BSimilar price

-29 vs GPT-5.5 (predecessor) · Ahead of 3 of 4 similarly priced models

Arena Search

Best Arena rating per model · Higher is better

Not ranked on Arena Search yet

Arena Document

Best Arena rating per model · Higher is better

Arena #25

  1. 1481
    GPT-5.5Predecessor
  2. 1467
    Claude Opus 4.5Nearby rank
  3. 1466
    Grok 4.5Nearby rank
  4. 1460
    GPT-5.6 Luna
  5. 1460
    GPT-5.5 InstantNearby rank
  6. 1459
    Gemini 3.1 Pro PreviewNearby rank
  7. 1452
    Gemma 4 31BSimilar price
  8. 1446
    Qwen3.7 PlusSimilar price
  9. 1435
    MiniMax M3Similar price

-21 vs GPT-5.5 (predecessor) · Ahead of 1 of 3 similarly priced models

Agent Arena

Best IPS score per model · Higher is better

Not ranked on Agent Arena yet

ARC-AGI-2

Best score per model · Higher is better

#38 of 85

  1. 85.0%
    GPT-5.5Predecessor
  2. 72.9%
    DeepSeek V4.1 FlashSimilar price
  3. 60.4%
    Gemini 3.6 FlashNearby rank
  4. 60.4%
    Kimi K3Nearby rank
  5. 59.5%
    GPT-5.6 Luna
  6. 59.3%
    GPT-6 LunaNearby rank
  7. 54.2%
    GPT-5.2 ProNearby rank
  8. 42.4%
    Qwen3.8 27BSimilar price
  9. 4.9%
    MiniMax M2.5Similar price
  10. 0.0%
    Llama 4 MaverickSimilar price

-25.5 pts vs GPT-5.5 (predecessor) · Ahead of 3 of 4 similarly priced models

FrontierMath-Tier-4-v2-Private

Best score per model · Higher is better

#18 of 72

  1. 72.5%
    GPT-5.5Predecessor
  2. 70.7%
    GPT-5.6 TerraNearby rank
  3. 61.0%
    GPT-5.6 Luna
  4. 58.5%
    GPT-5.4 ProNearby rank
  5. 56.1%
    Claude Opus 4.8Nearby rank
  6. 29.3%
    GLM 5.3Similar price
  7. 26.8%
    DeepSeek V4.1 FlashSimilar price
  8. 17.1%
    Inkling SmallSimilar price
  9. 0.0%
    Gemini 3.5 Flash LiteSimilar price

-11.5 pts vs GPT-5.5 (predecessor) · Ahead of 4 of 4 similarly priced models

FrontierMath-Tiers-1-3-v2-Private

Best score per model · Higher is better

#18 of 87

  1. 85.3%
    GPT-5.5Predecessor
  2. 82.5%
    GPT-5.4 ProNearby rank
  3. 82.1%
    GPT-5.6 Luna
  4. 80.0%
    Claude Opus 4.8Nearby rank
  5. 78.9%
    GPT-6 LunaNearby rank
  6. 68.8%
    GLM 5.3Similar price
  7. 67.4%
    DeepSeek V4.1 FlashSimilar price
  8. 27.7%
    Gemini 3.1 Flash LiteSimilar price
  9. 22.5%
    Qwen3.6 FlashSimilar price

-3.2 pts vs GPT-5.5 (predecessor) · Ahead of 4 of 4 similarly priced models

GPQA diamond

Best score per model · Higher is better

#29 of 161

  1. 91.9%
    GLM 5.2Nearby rank
  2. 91.7%
    DeepSeek V4 Pro 0813Nearby rank
  3. 91.6%
    GPT-5.6 Luna
  4. 91.4%
    GPT-5.2Nearby rank
  5. 91.0%
    Claude Opus 4.8Nearby rank
  6. 90.9%
    MiniMax M3Similar price
  7. 90.9%
    GLM 5.3Similar price
  8. 90.7%
    GPT-5.5Predecessor
  9. 83.3%
    Qwen3.6 FlashSimilar price
  10. 71.2%
    DeepSeek V3Similar price

+0.9 pts vs GPT-5.5 (predecessor) · Ahead of 4 of 4 similarly priced models

OTIS Mock AIME 2024-2025

Best score per model · Higher is better

#24 of 158

  1. 98.3%
    DeepSeek V4.1 FlashNearby rank
  2. 98.3%
    Mistral Large 4Nearby rank
  3. 98.3%
    GPT-5.6 Luna
  4. 98.1%
    Grok 4.7Nearby rank
  5. 97.8%
    Claude Opus 4.7Nearby rank
  6. 91.1%
    GLM 5.3Similar price
  7. 84.4%
    GPT-5.5Predecessor
  8. 84.4%
    Qwen3.6 FlashSimilar price
  9. 71.1%
    MiniMax M3Similar price
  10. 48.9%
    DeepSeek V3Similar price

+13.9 pts vs GPT-5.5 (predecessor) · Ahead of 4 of 4 similarly priced models

SimpleQA Verified

Best score per model · Higher is better

#50 of 87

  1. 63.0%
    GPT-5.5Predecessor
  2. 42.9%
    DeepSeek V4.1 FlashSimilar price
  3. 42.5%
    DeepSeek V4.1 Flash 1017Nearby rank
  4. 41.4%
    GPT-6 LunaNearby rank
  5. 41.0%
    GPT-5.6 Luna
  6. 41.0%
    GLM 5.3Similar price
  7. 40.4%
    Qwen3 235B A22B Thinking 2507Nearby rank
  8. 19.1%
    Inkling SmallSimilar price
  9. 15.9%
    Qwen3.6 FlashSimilar price

-22.0 pts vs GPT-5.5 (predecessor) · Ahead of 2 of 4 similarly priced models

DeepSWE 1.1

Primary coding-agent benchmark · 113 tasks · Data generated 2026-09-22

Official leaderboard
67.2%
pass@1 · highest published configuration
mini-swe-agent · max reasoning
pass@4
90.3%
Average cost / attempt
$3.03
Benchmark runs
4

Score vs cost by reasoning effort

Average cost per attempt (USD) against pass rate

All published configurations

Harness / reasoningpass@1pass@4Cost / attempt
mini-swe-agent · max67.2%90.3%$3.03
mini-swe-agent · xhigh56.9%78.8%$1.54
mini-swe-agent · high44.2%75.2%$0.78
mini-swe-agent · medium11.3%27.4%$0.22
mini-swe-agent · low1.5%4.4%$0.07

pass@1 is the success rate across scored attempts; pass@4 is the share of tasks solved in at least one of four attempts. Costs are reported by DeepSWE for each configuration. These results are not directly comparable to evaluations using different harnesses. Official source JSON.

Reasoning effort

ARC-AGI-2

Score by reasoning effort · Higher is better

FrontierMath-Tiers-1-3-v2-Private

Score by reasoning effort · Higher is better

GPQA diamond

Score by reasoning effort · Higher is better

OTIS Mock AIME 2024-2025

Score by reasoning effort · Higher is better

All published scores

Benchmark compilation: Epoch AI, Capabilities & benchmarking (CC BY attribution). Scores are converted to percentages and model names matched to the Vertical Tabs catalog. Individual evaluations are credited to their original sources. External results may carry separate licences and evaluation configurations; do not interpret scores from different harnesses as directly comparable.

Arena ratings: Arena (lmarena-ai/leaderboard-dataset), licensed CC BY 4.0. Ratings come from pairwise human votes and are only comparable within one leaderboard; the interval is Arena's 95% confidence interval. Model names are matched to the Vertical Tabs catalog.

To give you the best experience, we use cookies for personalization, insights, and performance. Learn more.