MATH 500

AI benchmark results and model performance

Type: percentage

Recorded Results

Average Score

86.24%

Score Range

69.02% - 98.20%

Leading Model (lowest score)

69.02% - Granite 3.0 8B Instruct

Scores Over Time

Individual benchmark scores plotted by date.

Models Using This Benchmark

Organisation	Model	Reported	Top Score	Info	Self Reported	Source
IBM	Granite 3.0 8B Instruct	-	69.02%	inferred high-confidence family alias from granite-3.3-8b-instruct (score=0.4911; benches=14)	Yes	Source
IBM	Granite 3.1 8B Instruct	-	69.02%	inferred high-confidence family alias from granite-3.3-8b-instruct (score=0.4911; benches=14)	Yes	Source
IBM	Granite 3.2 8B Instruct	-	69.02%	inferred high-confidence family alias from granite-3.3-8b-instruct (score=0.4911; benches=14)	Yes	Source
IBM	Granite 3.2 8B Instruct Preview	-	69.02%	inferred high-confidence family alias from granite-3.3-8b-instruct (score=0.4687; benches=14)	Yes	Source
IBM	Granite 3.3 2B Instruct	16 Apr 2025	69.02%	inferred family alias from granite-3.3-8b-instruct (score=0.3627; benches=14)	Yes	Source
IBM	Granite 3.3 8B Instruct	16 Apr 2025	69.02%	-	Yes	Source
IBM	Granite Guardian 3.0 8B	-	69.02%	inferred family alias from granite-3.3-8b-instruct (score=0.4062; benches=14)	Yes	Source
IBM	Granite Guardian 3.1 8B	-	69.02%	inferred family alias from granite-3.3-8b-instruct (score=0.4062; benches=14)	Yes	Source
IBM	Granite Guardian 3.3 8B	-	69.02%	inferred high-confidence family alias from granite-3.3-8b-instruct (score=0.5071; benches=14)	Yes	Source
IBM	Granite Speech 3.2 8B	-	69.02%	inferred family alias from granite-3.3-8b-instruct (score=0.4062; benches=14)	Yes	Source
IBM	Granite Speech 3.3 8B	-	69.02%	inferred high-confidence family alias from granite-3.3-8b-instruct (score=0.5071; benches=14)	Yes	Source
OpenAI	o1 mini	12 Sept 2024	90%	-	Yes	Source
DeepSeek	DeepSeek OCR	20 Oct 2025	90.20%	inferred family alias from deepseek-v3 (score=0.3000; benches=20)	Yes	Source
DeepSeek	DeepSeek V2 (2024-06-28)	28 Jun 2024	90.20%	inferred family alias from deepseek-v3 (score=0.4159; benches=20)	Yes	Source
DeepSeek	DeepSeek V4	-	90.20%	inferred high-confidence family alias from deepseek-v3 (score=0.5818; benches=20)	Yes	Source
Qwen	QwQ 32B	-	90.60%	-	Yes	Source
Qwen	QwQ 32B Preview	-	90.60%	-	Yes	Source
Microsoft	Phi 4 Mini Flash Reasoning	-	94.60%	inferred modality/version alias from phi-4-mini-reasoning	Yes	Source
Microsoft	Phi 4 Mini Reasoning	30 Apr 2025	94.60%	-	Yes	Source
Nvidia	Llama 3.1 Nemotron Nano 4B V1.1	-	95.40%	inferred high-confidence family alias from llama-3.1-nemotron-nano-8b-v1 (score=0.5523; benches=7)	Yes	Source
Nvidia	Llama 3.1 Nemotron Nano 8B V1	18 Mar 2025	95.40%	-	Yes	Source
MiniMax	Minimax M1 40K	16 Jun 2025	96%	-	Yes	-
Moonshot	Kimi K1.5	20 Jan 2025	96.20%	-	Yes	Source
Meituan	LongCat Flash Chat	-	96.40%	inferred high-confidence family alias from longcat-flash-chat (score=0.4667; benches=16)	Yes	Source
Nvidia	Llama 3.3 Nemotron Super 49B v1	18 Mar 2025	96.60%	-	Yes	Source
Nvidia	Llama 3.3 Nemotron Super 49B V1.5	-	96.60%	inferred version-family alias from llama-3.3-nemotron-super-49b-v1	Yes	Source
MiniMax	MiniMax M1 80K	16 Jun 2025	96.80%	-	Yes	-
Nvidia	Llama 3.1 Nemotron Ultra 253B v1	07 Apr 2025	97%	-	Yes	Source
Moonshot	Kimi K2 (2025-09-05)	05 Sept 2025	97.40%	Acc	Yes	Source
Nvidia	Nvidia Nemotron Nano 12B V2	-	97.80%	inferred high-confidence family alias from nvidia-nemotron-nano-9b-v2 (score=0.4889; benches=6)	Yes	Source
Nvidia	Nvidia Nemotron Nano 9B V2	-	97.80%	-	Yes	Source
z.AI	GLM 4.5 Air	28 Jul 2025	98.10%	-	Yes	Source
z.AI	GLM 4.5	28 Jul 2025	98.20%	-	Yes	Source