Benchmark Explorer

Explore how models perform on various benchmarks

Benchmarks

🔍

IOI

The International Olympiad in Informatics (IOI) benchmark evaluates models on competitive programming problems from the 2024 and 2025 IOI competitions. Models must write C++ solutions at the level of elite high-school programmers, with access to execution environments and submission tools matching actual competition conditions. Unlike the saturating IMO benchmark, IOI provides clear differentiation between model capabilities with standardized, automated grading.
Source:

Model Performance

#1
Grok-4
26.2%
#2
Grok-4-Thinking
26.2%
#3
Gemini-2.5-Pro-Thinking
17.1%
#4
Gemini-2.5-Pro-0325
17.1%
#5
Claude-4.1-Opus
15.2%
#6
Qwen-3-Max-Preview
7.8%
#7
Claude-4.0-Sonnet
6.5%
#8
OpenAI-O4-mini-medium
5.3%
#9
OpenAI-O4-mini-high
5.3%
#10
Claude-4.0-Sonnet-Thinking
4.6%
#11
Gemini-2.5-Flash
3.9%
#12
DeepSeek-V3-0324
1.7%
#13
Kimi-K2-Instruct
1.3%
#14
Qwen-3
0.0%
#15
Qwen-3-Thinking
0.0%