Skip to content

Benchmark

MGSM

Math benchmarks measure how well models solve maths problems. Scores are percentages. Higher is better.

Models and systems measured
31
Scores recorded
32
Sources
2
IQ weight
2.25

Score distribution

81%
Mean
86.4%
Median
11.8%
Std dev
58.7%
10th percentile
91.1%
90th percentile
53.1%
Min
53.1%92.3%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.25 for chat models, measured over 27 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model or systemScore
1Llama 4 Maverick· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim92.3%
2o3 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim92%
3Claude 3.5 Sonnet· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim91.6%
4Llama 3.3 70B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim91.1%
5O1 Preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim90.8%
6Claude 3 Opus· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim90.7%
7Llama 4 Scout· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim90.6%
8Llama 4 Scoutinstruction-tuned table· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.90.6%
9GPT 4o· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim90.5%
10o1· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim89.3%
11GPT-4 Turbo· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim88.5%
12Gemini 1.5 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim87.5%
13GPT-4o mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim87%
14Llama 3.1 70B Instructmultilingual average· Reported by Meta· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.86.9%
15Claude Haiku 3.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim85.6%
16Qwen 3 235B A22B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim83.5%
17Claude 3 Sonnet· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim83.5%
18Gemini 1.5 Flash· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim82.6%
19Phi 4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim80.6%
20Claude 3 Haiku· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim75.1%
21GPT 4· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim74.5%
22Llama 3.2 11B Instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim68.9%
23Gemma 3n E4B IT· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim67%
24Phi 4 Mini· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim63.9%
25phi-3.5-moe-instruct· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim58.7%
26Llama 3.2 3B· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim58.2%
27GPT-3.5 Turbo· LLM Stats data56.3%
28Gemma 3n E2b It· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim53.1%

Not ranked

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
DeepSeek V4.1 Flash BaseEM · 8-shot· Reported by DeepSeek· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.80.2%
Qwen3.7-Plus-BaseQwen3.8-Flash-Next release/model card · base-model table· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.85.4%
Qwen3.8-27B-BaseQwen3.8-Flash-Next release/model card · base-model table· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.86.4%
Qwen3.8-Flash-Next-BaseQwen3.8-Flash-Next release/model card · base-model table· Reported by Qwen· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.89.3%

Data credits: Data by LLM Stats · BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links.