Arena-Hard v2 Benchmark

Arena-Hard v2 Benchmark | Phaseo

Models Using This Benchmark

Organisation	Model	Reported	Top Score	Info	Self Reported	Source
Xiaomi	MiMo V2 Flash	16 Dec 2025	86.20%	-	Yes	Source
Qwen	Qwen 3 Next 80B A3B Instruct	-	82.70%	-	Yes	Source
Qwen	Qwen 3 235B A22B Thinking 2507	-	79.70%	-	Yes	Source
Qwen	Qwen 3 VL 235B A22B Instruct	-	77.40%	-	Yes	-
Nvidia	Nemotron 3 Super	11 Mar 2026	73.88%	-	Yes	Source
Nvidia	Nemotron Nano 3 30B A3B	15 Dec 2025	67.70%	-	Yes	Source
Qwen	Qwen 3 VL 32B Instruct	-	64.70%	-	Yes	-
Qwen	Qwen 3 Next 80B A3B Thinking	-	62.30%	-	Yes	Source
Qwen	Qwen 3 VL 32B Thinking	-	60.50%	-	Yes	-
Qwen	Qwen 3 VL 30B A3B Instruct	-	58.50%	-	Yes	-
Qwen	Qwen 3 VL 30B A3B Thinking	-	56.70%	-	Yes	-
Qwen	Qwen 3 8B	-	51.10%	inferred high-confidence family alias from qwen3-vl-8b-thinking (score=0.4600; benches=50)	Yes	-
Qwen	Qwen 3 Embedding 8B	-	51.10%	inferred family alias from qwen3-vl-8b-thinking (score=0.3850; benches=50)	Yes	-
Qwen	Qwen 3 Guard Gen 8B	-	51.10%	inferred family alias from qwen3-vl-8b-thinking (score=0.3400; benches=50)	Yes	-
Qwen	Qwen 3 Guard Stream 8B	-	51.10%	inferred family alias from qwen3-vl-8b-thinking (score=0.3371; benches=50)	Yes	-
Qwen	Qwen 3 Reranker 8B	-	51.10%	inferred family alias from qwen3-vl-8b-thinking (score=0.3850; benches=50)	Yes	-
Qwen	Qwen 3 VL 8B Thinking	-	51.10%	-	Yes	-
Qwen	Qwen 3 VL Embedding 8B	-	51.10%	inferred high-confidence family alias from qwen3-vl-8b-thinking (score=0.5232; benches=50)	Yes	-
Qwen	Qwen 3 VL Reranker 8B	-	51.10%	inferred high-confidence family alias from qwen3-vl-8b-thinking (score=0.5275; benches=50)	Yes	-
Qwen	Qwen 3 4B Instruct 2507	-	36.80%	inferred family alias from qwen3-vl-4b-thinking (score=0.3462; benches=48)	Yes	-
Qwen	Qwen 3 4B SafeRL	-	36.80%	inferred family alias from qwen3-vl-4b-thinking (score=0.3850; benches=48)	Yes	-
Qwen	Qwen 3 4B Thinking 2507	-	36.80%	inferred family alias from qwen3-vl-4b-thinking (score=0.3462; benches=48)	Yes	-
Qwen	Qwen 3 Embedding 4B	-	36.80%	inferred family alias from qwen3-vl-4b-thinking (score=0.3850; benches=48)	Yes	-
Qwen	Qwen 3 Guard Gen 4B	-	36.80%	inferred family alias from qwen3-vl-4b-thinking (score=0.3400; benches=48)	Yes	-
Qwen	Qwen 3 Guard Stream 4B	-	36.80%	inferred family alias from qwen3-vl-4b-thinking (score=0.3371; benches=48)	Yes	-
Qwen	Qwen 3 Reranker 4B	-	36.80%	inferred family alias from qwen3-vl-4b-thinking (score=0.3850; benches=48)	Yes	-
Qwen	Qwen 3 VL 4B Thinking	-	36.80%	-	Yes	-

Average Score

Score Range

Leading Model

Arena-Hard v2

Recorded Results

Average Score

Score Range

Leading Model

Models Using This Benchmark