- Context
- 1Mtokens
- Released
- 3mo ago
Cost/1M tokens
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
#12 of 28
- 74.1%
- 73.8%
- 73.6%
- 72.7%
- 69.9%
- 69.6%
- 69.0%
- 68.5%
- 67.5%
- 67.2%
- 65.5%Gemini 3.7 Flash
Arena Text
Best Arena rating per model · Higher is better
Arena #16
- 1491
Muse Spark 1.1Nearby rank - 1490Claude Opus 5Nearby rank
- 1489Gemini 3.7 Flash
- 1487Kimi K3Nearby rank
- 1487Gemini 3.1 Pro PreviewNearby rank
- 1483Gemini 3.6 FlashPredecessor
- 1457GLM 5Similar price
- 1448
Step 5 PreviewSimilar price - 1442Grok 4.3Similar price
- 1290
Nova Pro 1.0Similar price
Level with Gemini 3.6 Flash (predecessor, overlapping 95% CIs) · Ahead of 4 of 4 similarly priced models
Arena WebDev
Best Arena rating per model · Higher is better
Arena #40
- 1594Claude Haiku 5.5Nearby rank
- 1590
Qwen3.8 27BNearby rank - 1588Gemini 3.7 Flash
- 1585
MiniMax M3.1 FlashNearby rank - 1583Gemini 3.8 FlashNearby rank
- 1573Step 5 PreviewSimilar price
- 1538Gemini 3.6 FlashPredecessor
- 1435GLM 5Similar price
- 1400Qwen3.5 397B A17BSimilar price
- 1357Grok 4.3Similar price
+50 vs Gemini 3.6 Flash (predecessor) · Ahead of 4 of 4 similarly priced models
Arena Vision
Best Arena rating per model · Higher is better
Arena #8
- 1299Claude Opus 4.6Nearby rank
- 1295Claude Opus 5.5Nearby rank
- 1294Gemini 3.7 Flash
- 1294Muse Spark 1.2Nearby rank
- 1293Gemini 3.8 FlashNearby rank
- 1280Gemini 3.6 FlashPredecessor
- 1246Qwen3.5 397B A17BSimilar price
- 1245Step 5 PreviewSimilar price
- 1242Grok 4.3Similar price
- 1019Nova Pro 1.0Similar price
Level with Gemini 3.6 Flash (predecessor, overlapping 95% CIs) · Ahead of 4 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 #21
- 1480Muse Spark 1.1Similar price
- 1469Gemini 3.8 FlashNearby rank
- 1469Claude Sonnet 5Nearby rank
- 1468Gemini 3.7 Flash
- 1467Claude Opus 4.5Nearby rank
- 1466Grok 4.5Nearby rank
- 1454Gemini 3.6 FlashPredecessor
- 1450Kimi K2.6Similar price
- 1419Claude Haiku 4.5Similar price
- 1406GLM 5V TurboSimilar price
Level with Gemini 3.6 Flash (predecessor, overlapping 95% CIs) · Ahead of 3 of 4 similarly priced models
Agent Arena
Best IPS score per model · Higher is better
Arena #37
- +0.005
Hy4 previewSimilar price - +0.002GLM 5.2Similar price
- -0.002Grok 4.6Nearby rank
- -0.018GPT-5.6 TerraNearby rank
- -0.022Gemini 3.7 Flash
- -0.027MiniMax M3.1 FlashNearby rank
- -0.029GLM 5.3 FlashNearby rank
- -0.037Step 5 PreviewSimilar price
- -0.173
InklingSimilar price
Ahead of 1 of 4 similarly priced models
ARC-AGI-2
Best score per model · Higher is better
#14 of 85
- 88.3%Claude Opus 5Nearby rank
- 85.0%GPT-5.5Nearby rank
- 84.6%Gemini 3.7 Flash
- 84.6%GPT-5.5 ProNearby rank
- 83.9%GPT-5.6 TerraNearby rank
- 65.1%Grok 4.20Similar price
- 60.4%Gemini 3.6 FlashPredecessor
- 36.5%InklingSimilar price
- 18.9%GPT-5.4 MiniSimilar price
- 4.9%GLM 5Similar price
+24.2 pts vs Gemini 3.6 Flash (predecessor) · Ahead of 4 of 4 similarly priced models
FrontierMath-Tier-4-v2-Private
Best score per model · Higher is better
#29 of 72
- 46.0%GPT-5.2 ProNearby rank
- 39.0%Kimi K3Nearby rank
- 36.6%Gemini 3.7 Flash
- 34.1%Qwen3.7 MaxNearby rank
- 34.1%Qwen3.8 Max (0902)Nearby rank
- 29.3%GLM 5.2Similar price
- 25.6%Kimi K2.6Similar price
- 22.0%Gemini 3.6 FlashPredecessor
- 14.6%Grok 4.3Similar price
- 9.8%GPT-5.4 MiniSimilar price
+14.6 pts vs Gemini 3.6 Flash (predecessor) · Ahead of 4 of 4 similarly priced models
FrontierMath-Tiers-1-3-v2-Private
Best score per model · Higher is better
#27 of 87
- 74.0%GPT-5.2 ProNearby rank
- 72.2%Kimi K3Nearby rank
- 71.6%Gemini 3.7 Flash
- 71.2%
DeepSeek V4.1 Flash 1017Nearby rank - 70.2%Claude Opus 4.7Nearby rank
- 58.9%Gemini 3.6 FlashPredecessor
- 57.2%Kimi K2.6Similar price
- 51.2%GPT-5.4 MiniSimilar price
- 42.8%Grok 4.3Similar price
- 31.2%Qwen3.5 397B A17BSimilar price
+12.6 pts vs Gemini 3.6 Flash (predecessor) · Ahead of 4 of 4 similarly priced models
GPQA diamond
Best score per model · Higher is better
#7 of 161
- 95.4%Gemini 3.8 FlashNearby rank
- 95.4%GPT-6.1 SolNearby rank
- 94.8%Gemini 3.7 Flash
- 94.6%GPT-5.4 ProNearby rank
- 94.4%Gemini 3.1 Pro PreviewNearby rank
- 94.1%Gemini 3.6 FlashPredecessor
- 90.8%Kimi K2.6Similar price
- 88.8%Grok 4.3Similar price
- 87.8%GLM 5Similar price
- 86.4%Qwen3.5 397B A17BSimilar price
+0.7 pts vs Gemini 3.6 Flash (predecessor) · Ahead of 4 of 4 similarly priced models
OTIS Mock AIME 2024-2025
Best score per model · Higher is better
#32 of 158
- 97.8%GPT-5.4Nearby rank
- 97.8%Grok 4.5Nearby rank
- 97.2%Gemini 3.7 Flash
- 97.2%Kimi K3Nearby rank
- 96.7%DeepSeek V4 Pro 0423Nearby rank
- 96.1%Kimi K2.6Similar price
- 94.2%Gemini 3.6 FlashPredecessor
- 93.3%Grok 4.3Similar price
- 88.9%Qwen3.5 397B A17BSimilar price
- 80.0%GLM 5Similar price
+3.1 pts vs Gemini 3.6 Flash (predecessor) · Ahead of 4 of 4 similarly priced models
SimpleQA Verified
Best score per model · Higher is better
#12 of 87
- 69.7%Gemini 3.8 FlashNearby rank
- 69.7%GPT-5.6 SolNearby rank
- 69.2%Gemini 3.7 Flash
- 66.8%Gemini 3 Flash PreviewNearby rank
- 66.2%Gemini 3.5 FlashNearby rank
- 66.2%Gemini 3.6 FlashPredecessor
- 34.9%Kimi K2.6Similar price
- 34.2%GLM 5.2Similar price
- 33.2%Grok 4.3Similar price
- 29.4%GPT-5.4 MiniSimilar price
+3.0 pts vs Gemini 3.6 Flash (predecessor) · Ahead of 4 of 4 similarly priced models
DeepSWE 1.1
Primary coding-agent benchmark · 113 tasks · Data generated 2026-09-22
- pass@4
- 83.2%
- Average cost / attempt
- $2.03
- Benchmark runs
- 4
Score vs cost by reasoning effort
Average cost per attempt (USD) against pass rate
All published configurations
| Harness / reasoning | pass@1 | pass@4 | Cost / attempt |
|---|---|---|---|
| mini-swe-agent · medium | 65.5% | 83.2% | $2.03 |
| mini-swe-agent · high | 65.3% | 82.3% | $2.18 |
| mini-swe-agent · low | 53.8% | 77.9% | $1.83 |
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
All published scores
- Arena TextArena #161489
Level with Gemini 3.6 Flash (predecessor, overlapping 95% CIs)
- high148995% CI 1485 to 1494 · 33,300 votes
- Arena WebDevArena #401588
+50 vs Gemini 3.6 Flash (predecessor)
- high158895% CI 1582 to 1595 · 14,591 votes
- Arena VisionArena #81294
Level with Gemini 3.6 Flash (predecessor, overlapping 95% CIs)
- high129495% CI 1285 to 1302 · 6,840 votes
- Arena DocumentArena #211468
Level with Gemini 3.6 Flash (predecessor, overlapping 95% CIs)
- high146895% CI 1460 to 1476 · 7,680 votes
- Agent ArenaArena #37-0.022
- high-0.02295% CI -0.031 to -0.013 · 6,212,677 observations
- ARC-AGI-2#14 of 8584.6%
+24.2 pts vs Gemini 3.6 Flash (predecessor)
- low52.9%
- medium63.8%
- high84.6%
- FrontierMath-Tier-4-v2-Private#29 of 7236.6%
+14.6 pts vs Gemini 3.6 Flash (predecessor)
- high36.6%
Source:Epoch AI evaluation - FrontierMath-Tiers-1-3-v2-Private#27 of 8771.6%
+12.6 pts vs Gemini 3.6 Flash (predecessor)
- high71.6%
Source:Epoch AI evaluation - GPQA diamond#7 of 16194.8%
+0.7 pts vs Gemini 3.6 Flash (predecessor)
- high94.8%
Source:Epoch AI evaluation - OTIS Mock AIME 2024-2025#32 of 15897.2%
+3.1 pts vs Gemini 3.6 Flash (predecessor)
- high97.2%
Source:Epoch AI evaluation - SimpleQA Verified#12 of 8769.2%
+3.0 pts vs Gemini 3.6 Flash (predecessor)
- high69.2%
Source:Epoch AI evaluation
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.