DataLearner logo

Gemma 4 31BvsQwen3.5-27B

Across 10 shared benchmarks, Qwen3.5-27B leads overall: Gemma 4 31B wins 3, Qwen3.5-27B wins 7, with 0 ties and an average score difference of +20.45.

DeepMind
Gemma 4 31B

DeepMind · 2026-04-02 · Chat model

阿里巴巴
Qwen3.5-27B

阿里巴巴 · 2026-02-25 · Reasoning model

Gemma 4 31B3 wins(30%)(70%)7 winsQwen3.5-27B

Benchmark scores

Grouped by capability, sorted by largest gap within each. 10 shared benchmarks.

Agent Level Benchmark

Qwen3.5-27B 3/3
BenchmarkGemma 4 31BQwen3.5-27BDiff
τ²-Bench - Telecom65.50149 / 264Normal (With Tools)87.1074 / 264Normal (With Tools)-21.60
τ²-Bench76.9020 / 44Thinking (With Tools)7917 / 44Thinking (With Tools)-2.10
Terminal Bench Hard30.30110 / 244Normal (With Tools)31.80100 / 244Normal (With Tools)-1.50

General Knowledge

Qwen3.5-27B 2/2
BenchmarkGemma 4 31BQwen3.5-27BDiff
HLE11.80369 / 563Normal (No Tools) · Text only13.90348 / 563Normal (No Tools) · Text only-2.10
MMLU Pro85.2026 / 176Thinking (No Tools)86.1021 / 176Thinking (No Tools)-0.90

Coding and Software Engineer

Gemma 4 31B 1/1
BenchmarkGemma 4 31BQwen3.5-27BDiff
CodeForces2,15014 / 21Thinking (No Tools)1,89917 / 21Thinking (No Tools)+251

General Evaluation

Qwen3.5-27B 1/1
BenchmarkGemma 4 31BQwen3.5-27BDiff
GPQA Diamond76.30270 / 462Normal (No Tools)84.20174 / 462Normal (No Tools)-7.90

Instruction Following

Gemma 4 31B 1/1
BenchmarkGemma 4 31BQwen3.5-27BDiff
IF Bench53.50144 / 282Normal (No Tools)46.90171 / 282Normal (No Tools)+6.60

Long Context

Qwen3.5-27B 1/1
BenchmarkGemma 4 31BQwen3.5-27BDiff
AA-LCR46.70141 / 170Normal (No Tools)64122 / 170Normal (No Tools)-17.30

Multimodal Understanding

Gemma 4 31B 1/1
BenchmarkGemma 4 31BQwen3.5-27BDiff
MMMU-Pro70.30126 / 227Normal (No Tools)70130 / 227Normal (No Tools)+0.30

Specs

FieldGemma 4 31BQwen3.5-27B
PublisherDeepMind阿里巴巴
Release date2026-04-022026-02-25
Model typeChat modelReasoning model
ArchitectureDenseDense
Parameters30.7B27B
Context length256K1010K
Max output32K248320

Summary

  • Gemma 4 31Bleads in:Coding and Software Engineer (1/1), Instruction Following (1/1), Multimodal Understanding (1/1)
  • Qwen3.5-27Bleads in:Agent Level Benchmark (3/3), General Knowledge (2/2), General Evaluation (1/1), Long Context (1/1)

On average across the 10 shared benchmarks, Gemma 4 31B scores 20.45 higher.

Largest single-benchmark gap: CodeForces — Gemma 4 31B 2,150 vs Qwen3.5-27B 1,899 (+251).

Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.