Gemma 4 31BvsKimi K2.5
Across 11 shared benchmarks, Kimi K2.5 leads overall: Gemma 4 31B wins 3, Kimi K2.5 wins 8, with 0 ties and an average score difference of -2.32.
Gemma 4 31B
DeepMind · 2026-04-02 · Chat model
Kimi K2.5
Moonshot AI · 2026-01-27 · Multimodal model
Gemma 4 31B3 wins(27%)(73%)8 winsKimi K2.5
Benchmark scores
Grouped by capability, sorted by largest gap within each. 11 shared benchmarks.
Agent Level Benchmark
Even 2/2| Benchmark | Gemma 4 31B | Kimi K2.5 | Diff |
|---|---|---|---|
| τ²-Bench - Telecom | 65.50149 / 264Normal (With Tools) | 81.30108 / 264Normal (With Tools) | -15.80 |
| Terminal Bench Hard | 30.30110 / 244Normal (With Tools) | 18.90150 / 244Normal (With Tools) | +11.40 |
General Knowledge
Even 2/2| Benchmark | Gemma 4 31B | Kimi K2.5 | Diff |
|---|---|---|---|
| MMLU Pro | 85.2026 / 176Thinking (No Tools) | 78.5071 / 176Thinking (No Tools) | +6.70 |
| HLE | 11.80369 / 563Normal (No Tools) · Text only | 13.20357 / 563Normal (No Tools) · Text only | -1.40 |
Claw-style Agent Evaluation
Kimi K2.5 1/1| Benchmark | Gemma 4 31B | Kimi K2.5 | Diff |
|---|---|---|---|
| PinchBench v2 | 52.6638 / 45Reported best (effort unspecified) | 54.6036 / 45Reported best (effort unspecified) | -1.94 |
Coding and Software Engineer
Kimi K2.5 1/1| Benchmark | Gemma 4 31B | Kimi K2.5 | Diff |
|---|---|---|---|
| LiveCodeBench | 8052 / 250Thinking (No Tools) | 8529 / 250Thinking (No Tools) | -5 |
General Evaluation
Kimi K2.5 1/1| Benchmark | Gemma 4 31B | Kimi K2.5 | Diff |
|---|---|---|---|
| GPQA Diamond | 76.30270 / 462Normal (No Tools) | 78.90248 / 462Normal (No Tools) | -2.60 |
Instruction Following
Gemma 4 31B 1/1| Benchmark | Gemma 4 31B | Kimi K2.5 | Diff |
|---|---|---|---|
| IF Bench | 53.50144 / 282Normal (No Tools) | 43.70188 / 282Normal (No Tools) | +9.80 |
Long Context
Kimi K2.5 1/1| Benchmark | Gemma 4 31B | Kimi K2.5 | Diff |
|---|---|---|---|
| AA-LCR | 46.70141 / 170Normal (No Tools) | 67.30113 / 170Normal (No Tools) | -20.60 |
Math and Reasoning
Kimi K2.5 1/1| Benchmark | Gemma 4 31B | Kimi K2.5 | Diff |
|---|---|---|---|
| AIME 2026 | 89.2025 / 29Thinking (No Tools) | 92.5021 / 29Thinking (No Tools) | -3.30 |
Multimodal Understanding
Kimi K2.5 1/1| Benchmark | Gemma 4 31B | Kimi K2.5 | Diff |
|---|---|---|---|
| MMMU-Pro | 70.30126 / 227Normal (No Tools) | 73.10110 / 227Normal (No Tools) | -2.80 |
Specs
| Field | Gemma 4 31B | Kimi K2.5 |
|---|---|---|
| Publisher | DeepMind | Moonshot AI |
| Release date | 2026-04-02 | 2026-01-27 |
| Model type | Chat model | Multimodal model |
| Architecture | Dense | MoE |
| Parameters | 30.7B | 1T |
| Context length | 256K | 256K |
| Max output | 32K | 16K |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | Gemma 4 31B | Kimi K2.5 |
|---|---|---|
| Text input | Not public | $0.6 / 1M tokens |
| Text output | Not public | $3 / 1M tokens |
| Cache read | Not public | $0.1 / 1M tokens |
One or both models have incomplete public pricing.
Summary
- Gemma 4 31Bleads in:Instruction Following (1/1)
- Kimi K2.5leads in:Claw-style Agent Evaluation (1/1), Coding and Software Engineer (1/1), General Evaluation (1/1), Long Context (1/1), Math and Reasoning (1/1), Multimodal Understanding (1/1)
- Tied in:Agent Level Benchmark, General Knowledge
On average across the 11 shared benchmarks, Kimi K2.5 scores 2.32 higher.
Largest single-benchmark gap: AA-LCR — Gemma 4 31B 46.70 vs Kimi K2.5 67.30 (-20.60).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.