Qwen3.6-35B-A3B

Qwen · released 2026-04-01 · Apache-2.0 license

Qwen3.6-35B-A3B is Alibaba's April 2026 open-weight MoE with only ~3B active parameters, giving strong agentic coding (73.4% SWE-bench Verified) at very low inference cost, a notch behind the dense 27B.

Key specs

TypeLocal open-weight
Parameters
Architecture
Context window
Knowledge cutoff
Modalitiestext, vision
Recommended backends
Minimum viable rig

Benchmark scores

GPQA Diamond86%
SWE-bench Verified73.4%
AIME92.7%
MMLU-Pro85.2%
BFCL v3 (tool use)
Composite score7.12
Community ratingNo reviews yet

Strengths & weaknesses

Strengths: Sparse MoE: ~3B active, runs at ~3B inference cost with strong agentic coding; Solid SWE-bench Verified (73.4) for a tiny active footprint; Multimodal, Apache-2.0

Weaknesses: Trails the dense 27B sibling on most benchmarks; Low HLE (21.4) limits hardest reasoning