Note: Overall leaderboard rankings may not reflect true model quality — individual benchmarks give a clearer picture. ARC-Challenge MMLU GPQA GSM8K Artificial Analysis Intelligence Index v4.1
← Back to leaderboard

Microsoft: phi-3-mini-4k-instruct

3 benchmarks
CompanyMicrosoft
Country of originUSA

Leaderboard position

#931
Overall rank (of 1122 models)
128
Elo rating
3
Benchmarks
Top 83.0%
Percentile
#15
In Microsoft (of 19 models)
Above: Qwen: qwen2.5-1.5b #930 Below: fastino: gliner2.5-small-v1 #932

Benchmark highlights

Best benchmark
GSM8K 85.7
Worst benchmark
Artificial Analysis Intelligence Index v4.1 (Maximum Reasoning) 5

Nearest competitor from another company

The model from another company with the most similar benchmark scores.

appeared as nearest in 1 of 3 benchmarks

Intern Language Models: internlm2_5-7b-chat
1/3 33.3%
Qwen: qwen2.5-coder-7b-instruct
1/3 33.3%
Qwen: Qwen3.5 0.8B
1/3 33.3%
Model Votes % of benchmarks
Intern Language Models: internlm2_5-7b-chat 1 33.3%
Qwen: qwen2.5-coder-7b-instruct 1 33.3%
Qwen: Qwen3.5 0.8B 1 33.3%

Benchmarks

GSM8K 85.7 Artificial Analysis Intelligence Index v4.0 (Maximum Reasoning) 10.1 Artificial Analysis Intelligence Index v4.1 (Maximum Reasoning) 5

Predicted scores

Estimated from 25 similar models and the global Elo rating. These are projections, not measured results.

Benchmark Predicted Confidence Based on
MATH-500 ~73.81 60% Qwen: Qwen3 8B, Google: Gemma 3 4B Instruct, Qwen: qwen3-1.7b
AIME 2025 ~42.28 60% Qwen: Qwen3 8B, OpenAI: gpt-oss-20b, Qwen: qwen3-1.7b
GPQA Diamond ~42.13 60% Qwen: Qwen3 8B, Google: Gemma 3 4B Instruct, OpenAI: gpt-oss-20b
MMLU-Pro ~43.43 60% IBM Granite: Granite 4.1 8B, IBM Granite: granite-4.1-30b, DeepSeek: deepseek-v3
Humanity's Last Exam (no tools) ~4 60% Qwen: Qwen3 8B, OpenAI: gpt-oss-20b, DeepSeek: DeepSeek V4 Pro
IFEval ~72.37 60% Qwen: Qwen3 8B, Google: Gemma 3 4B Instruct, Qwen: qwen3-1.7b
IFBench ~41.34 60% Google: Gemma 3 4B Instruct, Xiaomi: MiMo-V2.5 Pro
HumanEval ~52.67 60% DeepSeek: deepseek-v2.5, Qwen: qwen3-1.7b
MBPP ~49.47 60% Qwen: qwen3-1.7b
Terminal-Bench v2.1 ~8.6 60% Xiaomi: MiMo-V2.5 Pro
MMLU ~48.99 60% Google: Gemma 3 4B Instruct, Qwen: Qwen3 0.6B, Google: Gemma 3 270M Instruct
SWE-bench Verified ~16.63 60% Qwen: Qwen3 8B, DeepSeek: deepseek-v2.5, OpenAI: gpt-oss-20b
SWE-bench Pro ~5.24 60% Meta Llama: Llama 3.1 405B Instruct, DeepSeek: DeepSeek V4 Pro, Xiaomi: MiMo-V2.5 Pro
SciCode ~15.47 60% Xiaomi: MiMo-V2.5 Pro
AA-LCR ~28.83 60% Xiaomi: MiMo-V2.5 Pro
Humanity's Last Exam (with tools) ~11.9 60% OpenAI: gpt-oss-20b
-(CoderPPL) ~-5.41 60% OpenAI: gpt-oss-20b, Qwen: Qwen3 0.6B
LiveCodeBench v6 ~39.37 60% OpenAI: gpt-oss-20b, Qwen: qwen3-1.7b, Xiaomi: MiMo-V2.5 Pro
MATH ~30.34 60% DeepSeek: deepseek-v2.5, OpenAI: o1, Xiaomi: MiMo-V2.5 Pro
BBH ~49.25 60% DeepSeek: deepseek-v2.5, Qwen: qwen3-1.7b, Xiaomi: MiMo-V2.5 Pro
ARC-Easy ~62.37 60% Google: Gemma 3 270M Instruct
ARC-Challenge ~48.31 60% Meta Llama: Llama 3.1 405B Instruct, Qwen: Qwen3 0.6B, Google: Gemma 3 270M Instruct
HellaSwag ~59.72 60% Qwen: Qwen3 0.6B, Google: Gemma 3 270M Instruct, Xiaomi: MiMo-V2.5 Pro
PIQA ~74.29 60% Google: Gemma 3 270M Instruct
Artificial Analysis Intelligence Index v4.3.2 ~5 60% Google: Gemma 3 270M Instruct