Qwen3-4B
4B hybrid-thinking dense model from the Qwen3 family, Apache 2.0.
About
Qwen3-4B is the 4-billion-parameter dense model in Alibaba's Qwen3 family, released April 2025 under Apache 2.0. Its defining feature is hybrid thinking: a single set of weights switches between step-by-step reasoning mode and fast direct-response mode via a flag, removing the need for separate reasoning and chat models. The Qwen team reports it can rival the much larger Qwen2.5-72B-Instruct despite its size.
Trained on 36 trillion tokens covering 119 languages, it supports tool and function calling, multilingual chat, and a 32K native context window (extendable). At roughly 2.5GB quantized it fits comfortably on consumer hardware, making it one of the strongest sub-5B open models for agentic and on-device use.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.
Use cases
Trust score — 8/10
Scored 8/10 (verified tier, metrics-v1). Evidence: Verified via vendor site/docs/GitHub and reputable coverage, wave 133; Trending score 84/100 from researcher assessment.
More in the qwen3 family
- Qwen3-Next-80B-A3B
- Qwen3-235B-A22B
- Qwen3-235B-A22B-2507
- Qwen3-4B-Instruct-2507
- Qwen3-8B
- Qwen3.5
- Qwen3-235B-A22B-Instruct-2507
- Qwen3 235B
- Qwen3-235B-A22B-Thinking-2507
- Qwen3-Omni-Flash
- Qwen3.6-27B
- Qwen3-32B
- Qwen3-Max
- Qwen3.8-Flash-Next
- Qwen3-Coder-Plus
- Qwen3
- Qwen3-0.6B
- Qwen3-Coder-Next
- Qwen3-4B-2507
- Qwen3-14B
- Qwen3-VL-30B-A3B
- Qwen3-1.7B
- Qwen3-Omni-30B-A3B