Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

SciCode

A scientific code generation benchmark that evaluates models' ability to solve realistic research problems across 16 subdomains in physics, mathematics, materials science, biology, and chemistry. Problems are derived from real research scenarios rather than exam-style questions.
Source:

Model Performance

#1
OpenAI-O4-mini-medium
47.0%
#2
OpenAI-O4-mini-high
47.0%
#3
Grok-4-Thinking
46.0%
#4
Grok-4
46.0%
#5
GPT-5
43.0%
#6
GPT-5-Thinking
43.0%
#7
Gemini-2.5-Pro-Thinking
43.0%
#8
Qwen-3-Thinking
42.0%
#9
Gemini-2.5-Pro-0605
42.0%
#10
Grok-3-Mini-Beta
41.0%
#11
GPT-5-mini
41.0%
#12
Claude-4.1-Opus-Thinking
41.0%
#13
Claude-4.0-Opus
41.0%
#14
OpenAI-O3-medium
41.0%
#15
OpenAI-O3-high
41.0%
#16
DeepSeek-R1
40.0%
#17
Claude-4.0-Opus-Thinking
40.0%
#18
OpenAI-O3-mini-high
40.0%
#19
OpenAI-O3-mini-medium
40.0%
#20
Claude-3.7-Sonnet-Thinking
40.0%
#21
Gemini-2.5-Pro-0325
40.0%
#22
Claude-4.0-Sonnet-Thinking
40.0%
#23
GPT-5-mini-Thinking
39.0%
#24
Gemini-2.5-Flash-Thinking
39.0%
#25
Claude-3.7-Sonnet
38.0%
#26
GPT-4.1
38.0%
#27
Claude-3.5-Sonnet-1022
37.0%
#28
GPT-5-nano-Thinking
37.0%
#29
Claude-4.0-Sonnet
37.0%
#30
Qwen-3-Max-Preview
37.0%
#31
Grok-3-Beta
37.0%
#32
Qwen-3
36.0%
#33
OpenAI-O1-1217
36.0%
#34
DeepSeek-V3-0324
36.0%
#35
GPT-OSS-120B
36.0%
#36
Kimi-K2-Instruct
35.0%
#37
DeepSeek-V3
35.0%
#38
GPT-5-nano
34.0%
#39
Mistral-Medium-3.1
34.0%
#40
Gemini-2.0-Flash
33.0%
#41
Gemini-2.0-Flash-Thinking-0121
33.0%
#42
Llama-4-Maverick-17B
33.0%
#43
OpenAI-O1-mini
32.0%
#44
Gemini-2.0-Pro-0121
31.0%
#45
GPT-4o-0513
31.0%
#46
Magistral-Medium-3.1
30.0%
#47
Llama-3.1-405B
30.0%
#48
Gemini-1.5-Pro
30.0%
#49
Mistral-Large-2
29.0%
#50
Gemini-2.5-Flash
29.0%
#51
Cohere-Command-A
28.0%
#52
Claude-3.5-Haiku
27.0%
#53
Llama-3.3-70B
26.0%
#54
Phi-4
26.0%
#55
GPT-4-mini
23.0%
#56
Claude-3-Sonnet
23.0%
#57
Claude-3-Opus
23.0%
#58
Mistral-Large
21.0%
#59
Cohere-Command-R-Plus
12.0%
#60
Llama-2-13B
12.0%
#61
Llama-2-7B
0.0%