Qwen3 4B 2507 (Reasoning)

Alibaba·Lançado em 2025-08-06

Open Source

Modelos Relacionados

Qwen: Qwen3.7 Plus2026-06-03 Qwen: Qwen3.7 Max2026-05-21 Qwen: Qwen3.6 27B2026-04-27 Qwen: Qwen3.6 35B A3B2026-04-27 Qwen: Qwen3.5 Plus 2026-04-202026-04-27 Qwen: Qwen3.6 Flash2026-04-27 Qwen: Qwen3.6 Max Preview2026-04-27 Qwen3.6 27B (Non-reasoning)2026-04-22

Índice de Qualidade

18.2

293rd de 533

Top 56%

Índice de Código

9.5

352nd de 443

Top 80%

Índice de Matemática

82.7

56th de 269

Top 21%

Posição no Mercado

Qwen3 4B 2507 (Reasoning)Média do Mercado

Benchmarks

MMLU-Pro

74.3%

GPQA Diamond

66.7%

HLE

5.9%

LiveCodeBench

64.1%

SciCode

25.6%

TerminalBench Hard

1.5%

MATH-500Não avaliado

AIMENão avaliado

AIME 2025

82.7%

IFBench

49.8%

Long Context Recall

37.7%

Tau2

25.4%

Média do MercadoMelhor Score

Open Source

Comparação Rápida

Modelos Similares

Magistral Small 1.2

Mistral

Q: 18.2$0.75/1M

Código: +5.3

Sarvam 105B (high)

Sarvam

Q: 18.2$0.07/1M

MiniCPM5-1B (Reasoning)

OpenBMB

Q: 18.2N/A/1M

Código: -8.0

Gemini 2.0 Pro Experimental (Feb '25)

Google

Q: 18.1N/A/1M

Código: +16.0

Meta: Llama 4 Maverick