Qwen3.5-27B

Qwen3.5 27B

Quick answer: Qwen3.5 27B is Alibaba's September 2025 dense 27B model scoring 61.0% BrowseComp, 48.5% HLE, and 76.5% IFBench. Apache 2.0 — notable for top HLE performance (48.5%) at the 27B dense scale.

At a Glance

Where Qwen3.5 27B leads

• 48.5% HLE — among the highest HLE for a 27B dense model; comparable to Qwen3.5 122B A10B (47.5%)
• 61.0% BrowseComp — strong web research
• 76.5% IFBench — solid instruction following
• Apache 2.0 — fully open
• Dense 27B: simpler deployment than MoE

Where it lags

• Limited benchmark coverage beyond 3 metrics
• 27B: moderate size vs larger frontier models

Best for: Open-source HLE-class reasoning at compact scale; research pipelines requiring high HLE at 27B parameters; cost-efficient frontier-reasoning deployment.

What Qwen3.5 27B Is

Qwen3.5 27B is the dense 27B model in Alibaba's Qwen3.5 generation, released September 2025. The Qwen3.5 generation showed significant HLE improvements over Qwen3 — this 27B model achieves 48.5% HLE, which is comparable to the 122B A10B MoE variant (47.5%) and above Gemini 3 Flash (43.5%).

This efficiency — frontier-class HLE at 27B dense — reflects Qwen3.5's improved reasoning training. The 61.0% BrowseComp further shows balanced multi-step research ability.

Specifications

Pricing

Open weights under Apache 2.0 — self-host at no cost.

Public Benchmark Scores

Qwen3.5 27B vs Alternatives

Qwen3.5 27B has near-equal or better HLE (48.5%) vs the 122B A10B MoE (47.5%) at a fraction of the total parameter count. For compact dense deployment with top HLE: Qwen3.5 27B.

Related Models & Benchmarks

• Compare: Qwen3.5 122B A10B · Kimi K2 Thinking 0905 · Qwen3 32B
• Benchmarks: BrowseComp · Humanity's Last Exam · IFBench

Frequently Asked Questions

Specs from Alibaba's Qwen3.5 27B release (September 2025) and Benchgen evaluations. Last updated 2026-07-24.