Model

Qwen3-14B (Thinking)

OPEN

Alibaba · Qwen3 family · released Apr 29, 2025

Mid-size Gen-3 dense model, thinking-mode setting; Apache 2.0.

ReasoningCodingVisionFunction callingTool useAgentic

1382.9

Elo · rank #156

Parameters

14.8B

Active params

14.8B (dense)

Context

128K tokens

Architecture

Dense Transformer (GQA, QK-Norm, hybrid thinking mode)

License

Apache 2.0

Languages

119+

API price (in/out)

No hosted API

Modalities

text

Benchmark results

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

AIMEMath79.3%#83

best: GPT-5.2 · 100.0%

Arena-HardHuman preference91.7%#9

best: Qwen3-235B-A22B (Non-Thinking) · 96.1%

BFCL v3Agents70.4%#8

best: Hunyuan-A13B · 78.3%

C-EvalKnowledge86.2%#22

best: Qwen3.6-Plus · 93.3%

CodeforcesCoding1766#28

best: DeepSeek-V4-Pro (Think Max) · 3206

GDPval-AAAgents500#49

best: Claude Fable 5 · 1932

GPQA DiamondReasoning64.0%#159

best: GPT-6 Astra · 96.0%

Humanity's Last ExamReasoning4.0%#139

best: Claude Opus 5 · 64.7%

IFBenchReasoning41.0%#51

best: MiniMax M3 · 83.0%

IFEvalReasoning85.4%#57

best: Gemma 4 26B A4B · 98.5%

LiveCodeBenchCoding63.5%#92

best: DeepSeek-V4-Pro (Think Max) · 93.5%

MATH-500Math96.8%#22

best: GPT-5 · 99.4%

MMLU-ReduxKnowledge88.6%#27

best: Qwen3.7-Max · 95.0%

MMMLUKnowledge77.9%#32

best: Gemini 3.1 Pro · 92.6%

Terminal-Bench 2.0Coding5.0%#76

best: Gemini 3.8 Flash · 89.4%

Run it locally

VRAM @ Q4

8.4 GB

VRAM @ FP16

29.6 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

~60 tok/s on RTX 4090 (Q4, llama.cpp)

Quantizations

GGUF Q4 · GGUF · FP8

Fine-tune it

Permissive

QLoRA11.3 GB1× RTX 3060 12GB

LoRA32.8 GB1× A100 80GB

Full fine-tune238.8 GB2× H200 141GB

QLoRA SFT on ~10k samples ≈ $8.66 (1× RTX 3060 12GB)

Plan a fine-tune →

Qwen3 family

Elo progression across releases

Qwen3-14B (Non-Thinking)Apr 20251060.2 Qwen3-14B (Thinking)Apr 20251382.9 Qwen3-235B-A22B (Non-Thinking)Apr 20251213.0 Qwen3-235B-A22B (Thinking)Apr 20251516.1 Qwen3-30B-A3B (Non-Thinking)Apr 20251070.4 Qwen3-30B-A3B (Thinking)Apr 20251452.3 Qwen3-32B (Non-Thinking)Apr 20251097.1 Qwen3-32B (Thinking)Apr 20251484.0 Qwen3-8B (Non-Thinking)Apr 2025900.6 Qwen3-8B (Thinking)Apr 20251311.4 Qwen3-235B-A22B-Instruct-2507Jul 20251559.3 Qwen3-235B-A22B-Thinking-2507Jul 20251827.4

Resources

HFQwen/Qwen3-14B GHQwenLM/Qwen3 DOCSqwen.readthedocs.io APIAvailable via Alibaba API

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