BenchPress - LLM Evaluation Leaderboard

Opinionated in scope. Objective in execution.

48 models · 80 prompts · 8 categories · Judges: gpt-4.1, claude-sonnet-4.6, gemini-2.5-flash, qwen3-235b · Updated Mar 07, 2026 06:56

Company Progress Over Time ?

Best Model per Company ?

Category Strengths by Company ?

Models by Company

Alibaba (3 models) - best: 0.93
Model Composite Judge DeepEval Avg Latency Efficiency
qwen3-235b 0.93 4.60/5 0.95 8.4s 0.48
qwen3-32b 0.86 4.22/5 0.92 4.3s 0.44
qwen3-coder-30b 0.84 4.12/5 0.90 1.9s 0.45
Amazon (4 models) - best: 0.84
Model Composite Judge DeepEval Avg Latency Efficiency
nova-2-lite 0.84 4.15/5 0.89 17.6s 0.41
nova-pro 0.78 3.79/5 0.85 9.7s 0.42
nova-lite 0.70 3.35/5 0.80 2.3s 0.37
nova-micro 0.68 3.35/5 0.78 14.1s 0.37
Anthropic (8 models) - best: 0.95
Model Composite Judge DeepEval Avg Latency Efficiency
claude-sonnet-4.6 0.95 4.78/5 0.96 19.2s 0.48
claude-opus-4.6 0.95 4.73/5 0.97 22.0s 0.47
claude-opus-4.5 0.94 4.73/5 0.96 20.7s 0.48
claude-sonnet-4.5 0.94 4.71/5 0.95 13.9s 0.50
claude-opus-4 0.93 4.62/5 0.95 16.5s 0.49
claude-sonnet-4 0.93 4.60/5 0.95 13.3s 0.48
claude-sonnet-3.7 0.87 4.31/5 0.92 7.3s 0.49
claude-haiku-3 0.71 3.48/5 0.80 3.9s 0.40
Cohere (1 model) - best: 0.83
Model Composite Judge DeepEval Avg Latency Efficiency
command-a 0.83 4.12/5 0.89 21.0s 0.45
Google (5 models) - best: 0.94
Model Composite Judge DeepEval Avg Latency Efficiency
gemini-3-flash 0.94 4.71/5 0.95 10.1s 0.49
gemini-3-pro 0.93 4.67/5 0.95 21.1s 0.48
gemini-3.1-pro 0.93 4.65/5 0.94 39.2s 0.48
gemini-2.5-flash 0.88 4.39/5 0.91 13.7s 0.44
gemma-3-27b 0.86 4.19/5 0.91 20.9s 0.42
Meta (5 models) - best: 0.83
Model Composite Judge DeepEval Avg Latency Efficiency
llama-4-maverick 0.83 4.07/5 0.88 1.4s 0.45
llama-4-scout 0.79 3.84/5 0.87 1.8s 0.43
llama3.1 0.67 3.27/5 0.77 15.6s 0.38
llama3.2-vision-11b 0.62 3.08/5 0.72 30.4s 0.35
llama3.2 0.59 2.95/5 0.70 13.5s 0.34
MiniMax (1 model) - best: 0.93
Model Composite Judge DeepEval Avg Latency Efficiency
minimax-m2.5 0.93 4.63/5 0.94 39.7s 0.44
Mistral (2 models) - best: 0.88
Model Composite Judge DeepEval Avg Latency Efficiency
mistral-large-3 0.88 4.34/5 0.93 24.4s 0.43
codestral 0.65 3.27/5 0.73 58.2s 0.38
Moonshot (1 model) - best: 0.94
Model Composite Judge DeepEval Avg Latency Efficiency
kimi-k2.5 0.94 4.67/5 0.95 41.1s 0.43
OpenAI (14 models) - best: 0.96
Model Composite Judge DeepEval Avg Latency Efficiency
gpt-5.3 0.96 4.83/5 0.97 5.7s 0.55
gpt-5.2 0.94 4.72/5 0.96 12.8s 0.50
gpt-5.4 0.94 4.70/5 0.96 13.7s 0.49
gpt-5.1 0.94 4.68/5 0.96 15.4s 0.49
gpt-4.1 0.92 4.55/5 0.95 9.0s 0.51
gpt-oss-120b 0.92 4.58/5 0.93 6.5s 0.41
o4-mini 0.91 4.52/5 0.95 13.0s 0.44
o3-mini 0.91 4.51/5 0.93 10.6s 0.45
gpt-4.1-mini 0.91 4.51/5 0.93 10.4s 0.50
gpt-oss-20b 0.88 4.43/5 0.90 142.1s 0.40
gpt-4.1-nano 0.87 4.26/5 0.92 5.1s 0.49
gpt-5 0.86 4.07/5 0.96 49.6s 0.36
gpt-4o 0.83 4.11/5 0.89 7.7s 0.48
gpt-4o-mini 0.82 4.01/5 0.88 7.8s 0.45
Zhipu (2 models) - best: 0.94
Model Composite Judge DeepEval Avg Latency Efficiency
glm-5 0.94 4.70/5 0.95 66.5s 0.42
glm-4.7-flash 0.86 4.26/5 0.90 38.6s 0.38
xAI (2 models) - best: 0.92
Model Composite Judge DeepEval Avg Latency Efficiency
grok-4 0.92 4.61/5 0.95 37.0s 0.47
grok-4.1-fast 0.90 4.44/5 0.94 9.4s 0.47