Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

Median Tokens/s

The median number of tokens processed per second, measuring the model's throughput in generating responses.
Source:

Model Performance

#1
GPT-OSS-120B
131.0%
#2
GPT-5-nano
129.9%
#3
Gemini-2.5-Flash-Thinking
124.9%
#4
OpenAI-O1-mini
115.5%
#5
Gemini-2.5-Flash
111.7%
#6
OpenAI-O3-medium
86.3%
#7
OpenAI-O3-high
86.3%
#8
Gemini-2.0-Flash
85.0%
#9
OpenAI-O1-1217
84.3%
#10
Gemini-2.5-Pro-0605
72.7%
#11
Gemini-2.5-Pro-0325
72.5%
#12
Gemini-2.5-Pro-Thinking
72.5%
#13
Magistral-Medium-3.1
72.4%
#14
GPT-5
71.5%
#15
OpenAI-O3-mini-high
71.3%
#16
Llama-4-Maverick-17B
70.5%
#17
OpenAI-O3-mini-medium
67.2%
#18
GPT-5-nano-Thinking
63.9%
#19
GPT-5-mini
61.2%
#20
Llama-2-7B
58.4%
#21
GPT-4.1
57.6%
#22
GPT-4o-0513
55.0%
#23
OpenAI-O4-mini-high
54.4%
#24
OpenAI-O4-mini-medium
54.4%
#25
Grok-3-Mini-Beta
50.8%
#26
Cohere-Command-A
49.5%
#27
GPT-3.5-Turbo
46.8%
#28
Llama-3.3-70B
39.1%
#29
Claude-3.7-Sonnet-Thinking
38.0%
#30
GPT-5-Thinking
37.1%
#31
Grok-4
36.7%
#32
Mistral-Medium-3.1
34.9%
#33
GPT-5-mini-Thinking
33.6%
#34
GPT-4-mini
33.5%
#35
Claude-3.7-Sonnet
32.6%
#36
Claude-3.5-Sonnet-1022
32.1%
#37
Claude-4.0-Sonnet
31.1%
#38
Qwen-3-Max-Preview
28.7%
#39
Claude-3.5-Haiku
27.4%
#40
Kimi-K2-Instruct
26.3%
#41
Qwen-3-Thinking
25.9%
#42
Claude-4.0-Opus-Thinking
25.5%
#43
Claude-4.0-Sonnet-Thinking
23.9%
#44
Mistral-Large
23.3%
#45
Claude-4.0-Opus
23.2%
#46
Grok-3-Beta
22.4%
#47
Claude-4.1-Opus
22.2%
#48
Claude-4.1-Opus-Thinking
21.4%
#49
Mistral-Large-2
21.2%
#50
Grok-4-Thinking
20.9%
#51
Qwen-3
18.1%
#52
Phi-4
17.1%
#53
Llama-3.1-405B
15.7%
#54
Gemini-2.0-Pro-0121
15.2%
#55
Claude-3-Opus
11.1%
#56
DeepSeek-R1
9.8%
#57
DeepSeek-V3-0324
9.7%
#58
Cohere-Command-R-Plus
8.4%
#59
Claude-3-Sonnet
0.0%
#60
Gemini-2.0-Flash-Thinking
0.0%
#61
Gemini-2.0-Flash-Thinking-0121
0.0%
#62
Gemini-1.5-Pro
0.0%
#63
GPT-4.5
0.0%
#64
DeepSeek-V3
0.0%
#65
Llama-2-13B
0.0%
#66
Grok-3-Think
0.0%
#67
Llama-2-70B
0.0%