- input
- $0.0434
- cached
- $0.035
- output
- $0.30
- Context
- 1Mtokens
- Released
- 3mo ago
- Size
- 304Bparameters
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
Not evaluated in DeepSWE 1.1
Arena Text
Best Arena rating per model · Higher is better
Not ranked on Arena Text yet
Arena WebDev
Best Arena rating per model · Higher is better
Not ranked on Arena WebDev yet
Arena Search
Best Arena rating per model · Higher is better
Not ranked on Arena Search yet
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
#31 of 85
- 65.1%
Grok 4.20Nearby rank - 64.6%
Claude Opus 4.6Nearby rank - 61.4%
DeepSeek V4 Flash 0731 - 61.4%Grok 4.7Nearby rank
- 61.3%Claude Haiku 5.5Similar price
- 2.6%
GPT-5 NanoSimilar price - 0.0%
Llama 4 ScoutSimilar price - 0.0%GPT-4.1 NanoSimilar price
Ahead of 4 of 4 similarly priced models
FrontierMath-Tier-4-v2-Private
Best score per model · Higher is better
#45 of 72
- 56.1%GPT-6 LunaSimilar price
- 46.3%Claude Haiku 5.5Similar price
- 26.8%
Mistral Large 4Nearby rank - 25.6%
Kimi K2.6Nearby rank - 24.4%DeepSeek V4 Flash 0731
- 24.4%Grok 4.5Nearby rank
- 22.0%
Gemini 3.6 FlashNearby rank - 2.4%GPT-5 NanoSimilar price
Ahead of 1 of 3 similarly priced models
FrontierMath-Tiers-1-3-v2-Private
Best score per model · Higher is better
#45 of 87
- 78.9%GPT-6 LunaSimilar price
- 75.1%Claude Haiku 5.5Similar price
- 58.9%Gemini 3.6 FlashNearby rank
- 58.6%Mistral Large 4Nearby rank
- 57.5%DeepSeek V4 Flash 0731
- 57.2%Kimi K2.6Nearby rank
- 57.2%Grok 4.5Nearby rank
- 20.0%GPT-5 NanoSimilar price
- 18.2%
Qwen3.5-FlashSimilar price
Ahead of 2 of 4 similarly priced models
GPQA diamond
Best score per model · Higher is better
#31 of 161
- 91.4%GPT-5.2Nearby rank
- 91.0%Claude Opus 4.8Nearby rank
- 91.0%DeepSeek V4 Flash 0731
- 90.9%DeepSeek V4 Pro 0423Nearby rank
- 90.9%
MiniMax M3Nearby rank - 82.3%Qwen3.5-FlashSimilar price
- 73.2%Gemma 4 26B A4BSimilar price
- 69.4%GPT-5 NanoSimilar price
- 56.1%
Phi 4Similar price
Ahead of 4 of 4 similarly priced models
OTIS Mock AIME 2024-2025
Best score per model · Higher is better
#43 of 158
- 94.7%Claude Sonnet 5Nearby rank
- 94.4%Claude Opus 4.6Nearby rank
- 94.4%DeepSeek V4 Flash 0731
- 94.2%Gemini 3.6 FlashNearby rank
- 93.9%
GLM 5.3 FlashNearby rank - 84.4%Qwen3.5-FlashSimilar price
- 82.2%Gemma 4 26B A4BSimilar price
- 81.1%GPT-5 NanoSimilar price
- 13.8%Phi 4Similar price
Ahead of 4 of 4 similarly priced models
SimpleQA Verified
Best score per model · Higher is better
#62 of 87
- 34.0%GLM 5.1Nearby rank
- 33.7%Claude Sonnet 5Nearby rank
- 33.6%DeepSeek V4 Flash 0731
- 33.2%Grok 4.3Nearby rank
- 32.2%GLM 4.7Nearby rank
- 23.8%Claude Haiku 5.5Similar price
- 20.3%Qwen3.5-FlashSimilar price
- 11.7%GPT-5 NanoSimilar price
- 6.0%GPT-4.1 NanoSimilar price
Ahead of 4 of 4 similarly priced models
Not evaluated in DeepSWE 1.1 as of 2026-09-22. Results are only shown for exact model versions in the official dataset. Official leaderboard.
Reasoning effort
ARC-AGI-2
Score by reasoning effort · Higher is better
All published scores
- ARC-AGI-2#31 of 8561.4%
- none2.1%
- low46.0%
- high56.0%
- max61.4%
- FrontierMath-Tier-4-v2-Private#45 of 7224.4%
- max24.4%
Source:Epoch AI evaluation - FrontierMath-Tiers-1-3-v2-Private#45 of 8757.5%
- max57.5%
Source:Epoch AI evaluation - GPQA diamond#31 of 16191.0%
- max91.0%
Source:Epoch AI evaluation - OTIS Mock AIME 2024-2025#43 of 15894.4%
- max94.4%
Source:Epoch AI evaluation - SimpleQA Verified#62 of 8733.6%
- max33.6%
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.