Model

Qwen2.5-7B

OPEN

Alibaba · Qwen2.5 family · released Sep 19, 2024

Highly-downloaded Apache-2.0 small dense model; strong MMLU-Pro/GSM8K for its size.

ReasoningCodingVisionFunction callingTool useAgentic

565.7

Elo · rank #285

Parameters

7.61B

Active params

7.61B (dense)

Context

128K tokens

Architecture

Dense transformer (GQA)

License

Apache-2.0

Languages

29+

API price (in/out)

No hosted API

Modalities

text

Benchmark results

Bar shows position within the tracked field; marker = field best

ARC-ChallengeReasoning63.7%#75

best: Llama 3.1 405B · 96.9%

BIG-Bench HardReasoning70.4%#56

best: ERNIE 4.5 300B-A47B · 94.3%

GPQA DiamondReasoning36.4%#235

best: GPT-5.6 · 94.6%

GSM8KMath85.4%#66

best: Llama 3.1 405B · 96.8%

HellaSwagReasoning80.2%#82

best: Claude 3 Opus · 95.4%

HumanEval+Coding50.6%#29

best: Mistral Small 3.2 (24B) · 92.9%

HumanEvalCoding57.9%#107

best: Claude Opus 4.5 · 99.4%

MATH-500Math49.8%#134

best: GPT-5 · 99.4%

MBPP+Coding62.9%#27

best: Llama 3.1 405B · 88.6%

MBPPCoding74.9%#25

best: Llama-3.3-Nemotron-Super-49B v1 (Reasoning On) · 91.3%

MMLU-ProKnowledge45.0%#143

best: Claude Fable 5 · 91.5%

MMLU-ReduxKnowledge71.1%#37

best: Qwen3.7-Max · 95.0%

MMLU (STEM)Knowledge72.3%#7

best: Falcon-H1 34B · 83.6%

MMLUKnowledge74.2%#126

best: OpenAI o3 · 92.9%

TruthfulQAKnowledge56.4%#34

best: Phi-3.5-MoE (16x3.8B, 6.6B active) · 77.5%

WinoGrandeReasoning75.9%#60

best: PaLM 2 · 90.9%

Run it locally

VRAM @ Q4

4.7 GB

VRAM @ FP16

15.2 GB

Fits on (Q4)

RTX 3060 12GBRTX 4070 Ti 16GBRTX 3090 24GBRTX 4090 24GBRTX 5090 32GBM4 Pro 48GBM3 Max 128GBM3 Ultra 512GBA100 80GBH100 80GBH200 141GBB200 192GB

Fits easily in 24GB VRAM at Q4 (~4.7GB); no independently verified 4090 llama.cpp tok/s figure found.

Quantizations

GGUF · AWQ · GPTQ Int4 · MLX

Qwen2.5 family

Elo progression across releases

Qwen2.5-14BSep 2024696.7 Qwen2.5-32BSep 2024843.7 Qwen2.5-72BSep 2024965.7 Qwen2.5-7BSep 2024565.7 Qwen2.5-MaxJan 20251252.5

Resources

HFQwen/Qwen2.5-7B GHQwenLM/Qwen2.5 DOCSqwenlm.github.io/blog/qwen2.5

API price weights · each benchmark row carries its own source badge (see methodology)