Skip to content

Benchmark

PinchBench

Coding benchmarks compare models on programming tasks. Scores are percentages.

Models measured
58
Scores recorded
58
Sources
2
IQ weight
1.13

Score distribution

75.1%
Mean
80.9%
Median
15.6%
Std dev
54.8%
10th percentile
89.7%
90th percentile
21.4%
Min
21.4%93.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 1.13 for chat models, measured over 58 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Claude Opus 4.8· PinchBench data93.5%
2Qwen3.7 Max· PinchBench data92.5%
3Muse Glimmer 30B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim90.5%
4Claude Opus 4.7· PinchBench data90.5%
5Nemotron 3 Ultra· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
90%
6MiMo V2.5· PinchBench data89.7%
7Claude Haiku 4.5· PinchBench data89%
8Grok Build 0.1· PinchBench data88.9%
9GPT-5.6 Luna· PinchBench data88.7%
10Qwen3.6 Flash· PinchBench data88.1%
11MiMo V2.5 Pro· PinchBench data87.5%
12Claude Opus 4.6· PinchBench data87.2%
13GLM 5.2· PinchBench data87%
14Nemotron 3.5 Lightning 30B A3B· PinchBench data
Values from 2 sources
86.4%
15Seed 2.0 Lite· PinchBench data85.3%
16GPT 5.5· PinchBench data84.4%
17Qwen3.6 Plus· PinchBench data84.2%
18GPT-5.6 Sol· PinchBench data84.2%
19ling-2.6-1t· PinchBench data82.6%
20Step 3.5 Flash· PinchBench data82.4%
21GLM 5V Turbo· PinchBench data
Values from 2 sources
82.2%
22DeepSeek V4 Flash· PinchBench data81.7%
23Gemini 3.1 Pro Preview· PinchBench data81.7%
24GPT 5.4· PinchBench data81.4%
25GPT-5.4 mini· PinchBench data81.3%
26MiMo V2 Omni· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim81.2%
27MiMo V2 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim81%
28Grok 4.20· PinchBench data80.9%
29Claude Sonnet 4.6· PinchBench data80.9%
30Gemini 3.1 Flash Lite· PinchBench data80.5%
31Devstral 2 123b Instruct 2512· PinchBench data80%
32GLM 5 Turbo· PinchBench data77.5%
33Mistral Small· PinchBench data76.3%
34Kimi K2.7 Code· PinchBench data76.1%
35Gemini 3 Flash· PinchBench data76%
36GPT-5.6 Terra· PinchBench data75.9%
37Kimi K2.5· PinchBench data75.6%
38Grok 4.5· PinchBench data75.2%
39Gemini 3.5 Flash· PinchBench data74.2%
40Gemma 4 26B A4B· PinchBench data72.6%
41Grok 4.3· PinchBench data71.7%
42GPT 5.4 Nano· PinchBench data71.4%
43Trinity Large Preview· PinchBench data69.3%
44LFM2.5 2.6B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim68.2%
45Trinity Large Thinking· PinchBench data67.4%
46Gemma 4 31B IT· PinchBench data65.7%
47MiniMax M2.7· PinchBench data65.6%
48Mistral Large 3· PinchBench data62.5%
49Aion 3.0· PinchBench data61.1%
50DeepSeek V4 Pro· PinchBench data61.1%

Showing 1–50 of 58 results.

Data credits: Data by LLM Stats