Gemma 4 26B A4BvsGemma 3 - 27B (IT)
Across 3 shared benchmarks, Gemma 4 26B A4B leads overall: Gemma 4 26B A4B wins 3, Gemma 3 - 27B (IT) wins 0, with 0 ties and an average score difference of +34.13.
Gemma 4 26B A4B
DeepMind · 2026-04-02 · Chat model
Gemma 3 - 27B (IT)
Google Deep Mind · 2025-03-12 · Chat model
Gemma 4 26B A4B3 wins(100%)(0%)0 winsGemma 3 - 27B (IT)
Benchmark scores
Grouped by capability, sorted by largest gap within each. 3 shared benchmarks.
Coding and Software Engineer
Gemma 4 26B A4B 1/1| Benchmark | Gemma 4 26B A4B | Gemma 3 - 27B (IT) | Diff |
|---|---|---|---|
| LiveCodeBench | 77.1036 / 126Thinking (No Tools) | 29.70122 / 126Normal (No Tools) | +47.40 |
General Evaluation
Gemma 4 26B A4B 1/1| Benchmark | Gemma 4 26B A4B | Gemma 3 - 27B (IT) | Diff |
|---|---|---|---|
| GPQA Diamond | 82.30108 / 226Thinking (No Tools) | 42.40208 / 226Normal (No Tools) | +39.90 |
General Knowledge
Gemma 4 26B A4B 1/1| Benchmark | Gemma 4 26B A4B | Gemma 3 - 27B (IT) | Diff |
|---|---|---|---|
| MMLU Pro | 82.6050 / 133Thinking (No Tools) | 67.50101 / 133Normal (No Tools) | +15.10 |
Specs
| Field | Gemma 4 26B A4B | Gemma 3 - 27B (IT) |
|---|---|---|
| Publisher | DeepMind | Google Deep Mind |
| Release date | 2026-04-02 | 2025-03-12 |
| Model type | Chat model | Chat model |
| Architecture | MoE | Dense |
| Parameters | 25.2B | 27B |
| Context length | 256K | 128K |
| Max output | 32K | Not available |
API pricing
Prices use DataLearner records when available; missing fields are not inferred.
| Item | Gemma 4 26B A4B | Gemma 3 - 27B (IT) |
|---|---|---|
| Text input | Not public | $0.09 / 1M tokens |
| Text output | Not public | $0.16 / 1M tokens |
One or both models have incomplete public pricing.
Summary
- Gemma 4 26B A4Bleads in:Coding and Software Engineer (1/1), General Evaluation (1/1), General Knowledge (1/1)
On average across the 3 shared benchmarks, Gemma 4 26B A4B scores 34.13 higher.
Largest single-benchmark gap: LiveCodeBench — Gemma 4 26B A4B 77.10 vs Gemma 3 - 27B (IT) 29.70 (+47.40).
Page generated from structured model, pricing and benchmark records. No real-time LLM is used to write the prose.