Qwen3 30B A3B
Qwen3's efficient MoE: 30B total, 3B active, Apache-2.0.
About
Qwen3-30B-A3B-Instruct-2507 is Alibaba's efficient mixture-of-experts model: 30.5B total parameters with only 3B active per token (128 experts, top-8 routing) — near-30B quality at 3B inference cost. 256K-token context (1M variant configs exist), hybrid thinking/non-thinking modes, 128K+ tokenizer. Apache-2.0 licensed, not gated, 998k HF downloads. It's the efficiency sweet spot of the Qwen3 generation: runs on a single high-end GPU while punching well above its active-parameter count.
Specs
Context window
Pricing
Free quota
API access
Weights
Released
Benchmarks
Provider-reported (Alibaba, Jul 2025): strong vs larger dense models on MMLU, HumanEval, MATH at 3B active — provider-reported.
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: HF downloads 998,093 (peer-normalized); released 442 days ago. Manual review: efficient MoE (30B total, 3B active, 256K), 998k dl; 2507 naming is confusing (multiple variants). License: Apache-2.0.
How the trust score is built
Full methodology →- Adoption · stars/downloads, log-scaled vs peers
- 8.8/10
- Maintenance · recency, releases, issue load
- 5.0/10
- Openness · license + docs
- 10.0/10
- Manual review · sanity read + red-flag check
- 9.0/10
- · HF downloads 998,093 (peer-normalized)
- · released 442 days ago
- · manual review: efficient MoE (30B total, 3B active, 256K), 998k dl; 2507 naming is confusing (multiple variants)