Qwen3.6-27BvsQwen3.5-27B
Across 13 shared benchmarks, Qwen3.6-27B leads overall: Qwen3.6-27B wins 10, Qwen3.5-27B wins 3, with 0 ties and an average score difference of +2.65.
Qwen3.6-27B
阿里巴巴 · 2026-04-22 · Reasoning model
Qwen3.5-27B
阿里巴巴 · 2026-02-25 · Reasoning model
Qwen3.6-27B10 wins(77%)(23%)3 winsQwen3.5-27B
Benchmark scores
Grouped by capability, sorted by largest gap within each. 13 shared benchmarks.
General Knowledge
Qwen3.6-27B 4/4| Benchmark | Qwen3.6-27B | Qwen3.5-27B | Diff |
|---|---|---|---|
| HLE | 15.10333 / 563Normal (No Tools) · Text only | 13.90348 / 563Normal (No Tools) · Text only | +1.20 |
| C-Eval | 91.405 / 48Thinking (No Tools) | 90.506 / 48Thinking (No Tools) | +0.90 |
| CritPt | 0.90158 / 200Normal (No Tools) | 0.30182 / 200Normal (No Tools) | +0.60 |
| MMLU Pro | 86.2019 / 176Thinking (No Tools) | 86.1021 / 176Thinking (No Tools) | +0.10 |
Agent Level Benchmark
Even 2/2| Benchmark | Qwen3.6-27B | Qwen3.5-27B | Diff |
|---|---|---|---|
| Terminal Bench Hard | 21.20145 / 244Normal (With Tools) | 31.80100 / 244Normal (With Tools) | -10.60 |
| τ²-Bench - Telecom | 93.6046 / 264Normal (With Tools) | 87.1074 / 264Normal (With Tools) | +6.50 |
AI Agent - Tool Usage
Qwen3.6-27B 1/1| Benchmark | Qwen3.6-27B | Qwen3.5-27B | Diff |
|---|---|---|---|
| Terminal Bench 2.0 | 59.3020 / 48Thinking (With Tools) | 41.6044 / 48Thinking (With Tools) | +17.70 |
Claw-style Agent Evaluation
Qwen3.5-27B 1/1| Benchmark | Qwen3.6-27B | Qwen3.5-27B | Diff |
|---|---|---|---|
| Claw Bench | 72.4027 / 29Thinking (With Tools) | 75.2026 / 29Thinking (With Tools) | -2.80 |
General Evaluation
Qwen3.6-27B 1/1| Benchmark | Qwen3.6-27B | Qwen3.5-27B | Diff |
|---|---|---|---|
| GPQA Diamond | 84.85159 / 462Normal (No Tools) | 84.20174 / 462Normal (No Tools) | +0.65 |
Instruction Following
Qwen3.5-27B 1/1| Benchmark | Qwen3.6-27B | Qwen3.5-27B | Diff |
|---|---|---|---|
| IF Bench | 45.70177 / 282Normal (No Tools) | 46.90171 / 282Normal (No Tools) | -1.20 |
Long Context
Qwen3.6-27B 1/1| Benchmark | Qwen3.6-27B | Qwen3.5-27B | Diff |
|---|---|---|---|
| AA-LCR | 66.70118 / 170Normal (No Tools) | 64122 / 170Normal (No Tools) | +2.70 |
Multimodal Understanding
Qwen3.6-27B 1/1| Benchmark | Qwen3.6-27B | Qwen3.5-27B | Diff |
|---|---|---|---|
| MMMU-Pro | 71.70118 / 227Normal (No Tools) | 70130 / 227Normal (No Tools) | +1.70 |
Text Embedding
Qwen3.6-27B 1/1| Benchmark | Qwen3.6-27B | Qwen3.5-27B | Diff |
|---|---|---|---|
| Context Arena | 53.7882 / 126Normal (No Tools) | 36.82102 / 126Normal (No Tools) | +16.96 |
Specs
| Field | Qwen3.6-27B | Qwen3.5-27B |
|---|---|---|
| Publisher | 阿里巴巴 | 阿里巴巴 |
| Release date | 2026-04-22 | 2026-02-25 |
| Model type | Reasoning model | Reasoning model |
| Architecture | Dense | Dense |
| Parameters | 27B | 27B |
| Context length | 128K | 1010K |
| Max output | 16K | 248320 |
Summary
- Qwen3.6-27Bleads in:General Knowledge (4/4), AI Agent - Tool Usage (1/1), General Evaluation (1/1), Long Context (1/1), Multimodal Understanding (1/1), Text Embedding (1/1)
- Qwen3.5-27Bleads in:Claw-style Agent Evaluation (1/1), Instruction Following (1/1)
- Tied in:Agent Level Benchmark
On average across the 13 shared benchmarks, Qwen3.6-27B scores 2.65 higher.
Largest single-benchmark gap: Terminal Bench 2.0 — Qwen3.6-27B 59.30 vs Qwen3.5-27B 41.60 (+17.70).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.