- input
- $2.00
- cached
- $0.50
- output
- $6.00
- Context
- 500Ktokens
- 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
#9 of 28
- 74.1%
- 73.8%
- 73.6%
- 72.7%
- 69.9%
- 69.6%
- 69.0%
- 68.5%
- 67.5%
Grok 4.6 - 67.2%
Arena Text
Best Arena rating per model · Higher is better
Arena #69
- 1483Kimi K3.1 ProSimilar price
- 1478Gemini 3.5 FlashSimilar price
- 1466Grok 4.5Predecessor
- 1457GPT-6 SolNearby rank
- 1457
Seed 2.0 ProNearby rank - 1456Grok 4.6
- 1456Claude Sonnet 4.5Nearby rank
- 1455GPT-5.1Similar price
- 1427
Mistral Medium 3.5Similar price
-9 vs Grok 4.5 (predecessor) · Ahead of 1 of 4 similarly priced models
Arena WebDev
Best Arena rating per model · Higher is better
Arena #32
- 1676
Qwen3.8 Max (0902)Similar price - 1622GLM 5.3Nearby rank
- 1621
DeepSeek V4.1 FlashNearby rank - 1618Grok 4.6
- 1615GPT-5.6 SolNearby rank
- 1609GLM 5.3 FlashNearby rank
- 1552Grok 4.5Predecessor
- 1500Gemini 3.5 FlashSimilar price
- 1417GPT-5Similar price
- 1263Mistral Medium 3.5Similar price
+66 vs Grok 4.5 (predecessor) · Ahead of 3 of 4 similarly priced models
Arena Vision
Best Arena rating per model · Higher is better
Arena #41
- 1281Gemini 3.5 FlashSimilar price
- 1280Grok 4.5Predecessor
- 1265Gemma 4 31BNearby rank
- 1265Kimi K2.6Nearby rank
- 1263Grok 4.6
- 1263DeepSeek V4.1 FlashNearby rank
- 1263Claude Sonnet 5.5Similar price
- 1251GPT-5.1Similar price
- 1199Mistral Medium 3.5Similar price
-17 vs Grok 4.5 (predecessor) · Ahead of 1 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 #34
- 1480
Muse Spark 1.1Similar price - 1469Claude Sonnet 5Similar price
- 1466Grok 4.5Predecessor
- 1459Gemini 3.5 FlashSimilar price
- 1451Gemini 3 Pro PreviewNearby rank
- 1450Kimi K2.6Nearby rank
- 1450Grok 4.6
- 1448Claude Sonnet 4.5Nearby rank
- 1446Qwen3.7 PlusNearby rank
- 1416GPT-5.1Similar price
-16 vs Grok 4.5 (predecessor) · Ahead of 1 of 4 similarly priced models
Agent Arena
Best IPS score per model · Higher is better
Arena #34
- +0.170Gemini 4 CarbonSimilar price
- +0.160Claude Sonnet 5.5Similar price
- +0.049Kimi K3.1 ProSimilar price
- +0.001
Qwen3.8 MaxNearby rank - -0.000
MiMo-V2.6-FlashNearby rank - -0.002Grok 4.6
- -0.018GPT-5.6 TerraNearby rank
- -0.022Gemini 3.7 FlashNearby rank
- -0.137Mistral Medium 3.5Similar price
Ahead of 1 of 4 similarly priced models
ARC-AGI-2
Best score per model · Higher is better
#27 of 85
- 83.9%Claude Sonnet 5.5Similar price
- 72.1%Claude Opus 4.8Nearby rank
- 72.1%Gemini 3.5 FlashSimilar price
- 67.1%Grok 4.6
- 65.8%GLM 5.3 FlashNearby rank
- 65.1%Grok 4.20Nearby rank
- 60.4%Kimi K3Similar price
- 52.6%Grok 4.5Predecessor
- 17.6%GPT-5.1Similar price
+14.4 pts vs Grok 4.5 (predecessor) · Ahead of 2 of 4 similarly priced models
FrontierMath-Tier-4-v2-Private
Best score per model · Higher is better
#32 of 72
- 80.5%Claude Sonnet 5.5Similar price
- 34.1%Qwen3.8 Max (0902)Similar price
- 31.7%Claude Opus 4.7Nearby rank
- 31.7%Grok 4.6
- 31.7%GPT-5.2Nearby rank
- 29.3%Claude Sonnet 5Nearby rank
- 26.8%Gemini 3.5 FlashSimilar price
- 24.4%Grok 4.5Predecessor
- 22.0%GPT-5Similar price
+7.3 pts vs Grok 4.5 (predecessor) · Ahead of 2 of 4 similarly priced models
FrontierMath-Tiers-1-3-v2-Private
Best score per model · Higher is better
#34 of 87
- 88.8%Claude Sonnet 5.5Similar price
- 67.4%DeepSeek V4.1 FlashNearby rank
- 66.0%Claude Opus 4.6Nearby rank
- 66.0%Grok 4.6
- 65.6%Claude Sonnet 5Nearby rank
- 65.6%Qwen3.8 Max (0902)Similar price
- 62.8%Gemini 3.5 FlashSimilar price
- 57.2%Grok 4.5Predecessor
- 55.4%GPT-5Similar price
+8.8 pts vs Grok 4.5 (predecessor) · Ahead of 3 of 4 similarly priced models
GPQA diamond
Best score per model · Higher is better
#13 of 161
- 94.3%GPT-6.1 TerraNearby rank
- 94.1%Gemini 3.6 FlashNearby rank
- 94.0%Grok 4.6
- 93.9%Claude Fable 5.2Nearby rank
- 93.9%Claude Opus 5Nearby rank
- 93.4%Grok 4.5Predecessor
- 92.8%Gemini 3.5 FlashSimilar price
- 92.3%Qwen3.8 Max (0902)Similar price
- 87.6%GPT-5.1Similar price
- 49.0%Mistral Large 2407Similar price
+0.6 pts vs Grok 4.5 (predecessor) · Ahead of 4 of 4 similarly priced models
OTIS Mock AIME 2024-2025
Best score per model · Higher is better
#17 of 158
- 100.0%Qwen3.8 Max (0902)Similar price
- 99.4%DeepSeek V4.1 Flash 1017Nearby rank
- 99.2%Muse Spark 1.3Nearby rank
- 99.2%Grok 4.6
- 98.9%Claude Haiku 5.5Nearby rank
- 98.9%Claude Opus 5Nearby rank
- 97.8%Grok 4.5Predecessor
- 95.6%Gemini 3.5 FlashSimilar price
- 91.4%GPT-5Similar price
- 8.5%Mistral Large 2407Similar price
+1.4 pts vs Grok 4.5 (predecessor) · Ahead of 3 of 4 similarly priced models
SimpleQA Verified
Best score per model · Higher is better
#32 of 87
- 66.2%Gemini 3.5 FlashSimilar price
- 50.1%GPT-5Similar price
- 49.4%o3Nearby rank
- 49.3%Grok 4.6
- 48.7%Qwen3 MaxNearby rank
- 48.3%Grok 4.5Predecessor
- 47.3%Qwen3.8 Max (0902)Similar price
- 46.5%Claude Sonnet 5.5Similar price
+1.0 pts vs Grok 4.5 (predecessor) · Ahead of 2 of 4 similarly priced models
DeepSWE 1.1
Primary coding-agent benchmark · 113 tasks · Data generated 2026-09-22
- pass@4
- 84.1%
- Average cost / attempt
- $3.45
- 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 | 67.5% | 84.1% | $3.45 |
| mini-swe-agent · xhigh | 66.7% | 85.0% | $5.50 |
| mini-swe-agent · high | 65.2% | 85.0% | $4.38 |
| mini-swe-agent · low | 41.6% | 69.0% | $1.04 |
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
GPQA diamond
Score by reasoning effort · Higher is better
OTIS Mock AIME 2024-2025
Score by reasoning effort · Higher is better
SimpleQA Verified
Score by reasoning effort · Higher is better
All published scores
- Arena TextArena #691456
-9 vs Grok 4.5 (predecessor)
- high145695% CI 1452 to 1461 · 34,616 votes
- Arena WebDevArena #321618
+66 vs Grok 4.5 (predecessor)
- high161895% CI 1611 to 1625 · 12,193 votes
- Arena VisionArena #411263
-17 vs Grok 4.5 (predecessor)
- high126395% CI 1255 to 1271 · 8,666 votes
- Arena DocumentArena #341450
-16 vs Grok 4.5 (predecessor)
- high145095% CI 1443 to 1458 · 7,669 votes
- Agent ArenaArena #34-0.002
- xhigh-0.00295% CI -0.014 to +0.010 · 3,806,387 observations
- ARC-AGI-2#27 of 8567.1%
+14.4 pts vs Grok 4.5 (predecessor)
- low27.6%
- medium61.3%
- high65.1%
- xhigh67.1%
- FrontierMath-Tier-4-v2-Private#32 of 7231.7%
+7.3 pts vs Grok 4.5 (predecessor)
- xhigh31.7%
Source:Epoch AI evaluation - FrontierMath-Tiers-1-3-v2-Private#34 of 8766.0%
+8.8 pts vs Grok 4.5 (predecessor)
- xhigh66.0%
Source:Epoch AI evaluation - GPQA diamond#13 of 16194.0%
+0.6 pts vs Grok 4.5 (predecessor)
- high94.0%
- xhigh93.2%
Source:Epoch AI evaluation - OTIS Mock AIME 2024-2025#17 of 15899.2%
+1.4 pts vs Grok 4.5 (predecessor)
- high97.8%
- xhigh99.2%
Source:Epoch AI evaluation - SimpleQA Verified#32 of 8749.3%
+1.0 pts vs Grok 4.5 (predecessor)
- high49.3%
- xhigh48.9%
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.