HMMT 2025 Benchmark

HMMT 2025 Benchmark | Phaseo

Models Using This Benchmark

Organisation	Model	Reported	Top Score	Info	Self Reported	Source
OpenAI	GPT 5.2 Pro	11 Dec 2025	100%	Feb 2025, No Tools	Yes	Source
OpenAI	GPT 5.2	11 Dec 2025	99.40%	Feb 2025, No Tools	Yes	Source
OpenAI	GPT 5.2 Chat	11 Dec 2025	99.40%	inferred alias from gpt-5.2-2025-12-11	Yes	Source
DeepSeek	DeepSeek V3.2 Speciale	01 Dec 2025	99.20%	-	Yes	Source
Moonshot	Kimi K2 Thinking	06 Nov 2025	97.50%	inferred alias from kimi-k2-thinking-0905	Yes	Source
OpenAI	GPT 5	07 Aug 2025	96.70%	Pass @ 1	Yes	Source
Qwen	Qwen 3.6 Plus	01 Apr 2026	96.70%	-	Yes	Source
SpaceXAI	Grok 4 Heavy	10 Jul 2025	96.70%	-	Yes	Source
Moonshot	Kimi K2.5	27 Jan 2026	95.40%	-	Yes	Source
Qwen	Qwen 3.5 397B A17B	16 Feb 2026	94.80%	HMMT Feb 25	Yes	Source
Nvidia	Nemotron 3 Super	11 Mar 2026	94.73%	-	Yes	Source
z.AI	GLM 5.1	07 Apr 2026	94%	Nov. 2025	Yes	Source
SpaceXAI	Grok 4	10 Jul 2025	93.90%	With Python	Yes	Source
Qwen	Qwen 3.5 27B	24 Feb 2026	92%	-	Yes	Source
Qwen	Qwen 3.5 Flash	23 Feb 2026	92%	inferred family alias from qwen3.5-27b (score=0.4147; benches=81)	Yes	Source
Qwen	Qwen 3.5 122B A10B	24 Feb 2026	91.40%	-	Yes	Source
Qwen	Qwen 3.5 35B A3B	24 Feb 2026	89.20%	-	Yes	Source
OpenAI	GPT 5 Mini	07 Aug 2025	87.80%	High Reasoning Effort, No Tools	Yes	Source
Xiaomi	MiMo V2 Flash	16 Dec 2025	84.40%	-	Yes	Source
Qwen	Qwen 3 235B A22B Thinking 2507	-	83.90%	-	Yes	Source
DeepSeek	DeepSeek OCR 2	-	83.60%	inferred family alias from deepseek-v3.2-exp (score=0.3809; benches=14)	Yes	Source
DeepSeek	DeepSeek V3.2 Exp	29 Sept 2025	83.60%	-	Yes	Source
Qwen	Qwen 3.5 9B	02 Mar 2026	83.20%	-	Yes	Source
Qwen	Qwen 3 VL 235B A22B Thinking	-	77.40%	-	Yes	-
Qwen	Qwen 3.5 4B	02 Mar 2026	76.80%	-	Yes	Source
OpenAI	GPT 5 Nano	07 Aug 2025	75.60%	High Reasoning Effort, No Tools	Yes	Source
Qwen	Qwen 3 Next 80B A3B Thinking	-	73.90%	-	Yes	Source
LG	EXAONE 4.0 32B	15 Jul 2025	72.90%	Reasoning	Yes	Source
Qwen	Qwen 3 VL 30B A3B Thinking	-	67.60%	-	Yes	-
Mistral	Ministral 3.0 14B	02 Dec 2025	67.50%	-	Yes	Source
Qwen	Qwen 3 8B	-	60.60%	inferred high-confidence family alias from qwen3-vl-8b-thinking (score=0.4600; benches=50)	Yes	-
Qwen	Qwen 3 Embedding 8B	-	60.60%	inferred family alias from qwen3-vl-8b-thinking (score=0.3850; benches=50)	Yes	-
Qwen	Qwen 3 Guard Gen 8B	-	60.60%	inferred family alias from qwen3-vl-8b-thinking (score=0.3400; benches=50)	Yes	-
Qwen	Qwen 3 Guard Stream 8B	-	60.60%	inferred family alias from qwen3-vl-8b-thinking (score=0.3371; benches=50)	Yes	-
Qwen	Qwen 3 Reranker 8B	-	60.60%	inferred family alias from qwen3-vl-8b-thinking (score=0.3850; benches=50)	Yes	-
Qwen	Qwen 3 VL 8B Thinking	-	60.60%	-	Yes	-
Qwen	Qwen 3 VL Embedding 8B	-	60.60%	inferred high-confidence family alias from qwen3-vl-8b-thinking (score=0.5232; benches=50)	Yes	-
Qwen	Qwen 3 VL Reranker 8B	-	60.60%	inferred high-confidence family alias from qwen3-vl-8b-thinking (score=0.5275; benches=50)	Yes	-
Qwen	Qwen 3 VL 235B A22B Instruct	-	57.40%	-	Yes	-
Mistral	Ministral 3.0 8B	02 Dec 2025	55.80%	-	Yes	Source
Qwen	Qwen 3 A235 A22B Instruct 2507	-	55.40%	-	Yes	Source
Qwen	Qwen 3 Next 80B A3B Instruct	-	54.10%	-	Yes	Source
Qwen	Qwen 3 4B Instruct 2507	-	53.10%	inferred family alias from qwen3-vl-4b-thinking (score=0.3462; benches=48)	Yes	-
Qwen	Qwen 3 4B SafeRL	-	53.10%	inferred family alias from qwen3-vl-4b-thinking (score=0.3850; benches=48)	Yes	-
Qwen	Qwen 3 4B Thinking 2507	-	53.10%	inferred family alias from qwen3-vl-4b-thinking (score=0.3462; benches=48)	Yes	-
Qwen	Qwen 3 Embedding 4B	-	53.10%	inferred family alias from qwen3-vl-4b-thinking (score=0.3850; benches=48)	Yes	-
Qwen	Qwen 3 Guard Gen 4B	-	53.10%	inferred family alias from qwen3-vl-4b-thinking (score=0.3400; benches=48)	Yes	-
Qwen	Qwen 3 Guard Stream 4B	-	53.10%	inferred family alias from qwen3-vl-4b-thinking (score=0.3371; benches=48)	Yes	-
Qwen	Qwen 3 Reranker 4B	-	53.10%	inferred family alias from qwen3-vl-4b-thinking (score=0.3850; benches=48)	Yes	-
Qwen	Qwen 3 VL 4B Thinking	-	53.10%	-	Yes	-
Mistral	Ministral 3.0 3B	02 Dec 2025	51.70%	-	Yes	Source
Qwen	Qwen 3 VL 30B A3B Instruct	-	50.60%	-	Yes	-
Moonshot	Kimi K2 (2025-09-05)	05 Sept 2025	38.80%	Avg@32	Yes	Source
LG	EXAONE 4.0 1.2B	15 Jul 2025	34%	Reasoning	Yes	Source
DeepSeek	DeepSeek V3.1	21 Aug 2025	33.50%	Non-thinking: 33.5%, Thinking: 84.2%	Yes	Source
DeepSeek	DeepSeek V3.1 Terminus	22 Sept 2025	33.50%	inferred alias from deepseek-v3.1	Yes	Source
Qwen	Qwen 3 VL 8B Instruct	-	32.50%	-	Yes	-
Qwen	Qwen 3 VL 4B Instruct	-	30.70%	-	Yes	-

Average Score

Score Range

Leading Model

HMMT 2025

Recorded Results

Average Score

Score Range

Leading Model

Models Using This Benchmark