See key specs and per-benchmark scores for each model/mode. Scroll horizontally for all columns. 当前对比 2 个模型的评测数据与核心参数。
Gemma 4 26B A4B
DeepMind
Each axis is a category average, normalized to a 100-point radar.
Relative edge: none clear / Relative gap: 编程与软件工程 -47.4
Relative edge: 编程与软件工程 +47.4 / Relative gap: none clear
Method: for each model and benchmark, the chart first averages all scores in the current mode scope instead of taking the best score, then averages those benchmark scores within each category. Only benchmarks with at least two selected models scored are included; missing values are not counted as zero.
Best overall
Gemma 4 26B A4B · 80.67
Best single
Gemma 4 26B A4B · MMLU Pro 82.60
Modality coverage
Gemma 3 - 27B (IT) · 2 modalities
Head to head
3
Benchmarks
0
Wins
3
Losses
-34.13
Average diff
Compare benchmark results across thinking modes and tool usage.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
Complete scores for each model/mode across selected benchmarks.
3 benchmarks with comparable scores. Each model shows its best score; mode label is displayed below.
| Benchmark | Gemma 3 - 27B (IT) | Gemma 4 26B A4B |
|---|---|---|
MMLU Pro 综合评估 | 67.50Standard Mode | 82.60Thinking Enabled |
LiveCodeBench 编程与软件工程 | 29.70Standard Mode | 77.10Thinking Enabled |
GPQA Diamond 科学与综合推理 | 42.40Standard Mode | 82.30Thinking Enabled |
Side-by-side input/output token pricing
Licensing, MoE architecture, and multi-modality support.
| Features & specs | Gemma 3 - 27B (IT)Google Deep Mind | Gemma 4 26B A4BDeepMind |
|---|---|---|
Core specsRelease | 2025-03-12 | 2026-04-02 |
Context length | 128K | 256K |
Parameters | 27B | 25.2B |
Active parameters | Not provided | 3.8B |
Max output | Not provided | 32768 |
MoE | No | Yes |
LicenseCode Open Source | Open Source · Gemma Terms of Use | Open Source · Apache 2.0 |
Weights Open Source | Open Source · Gemma Terms of Use | Open Source · Apache 2.0 |
Commercial use | 免费商用授权 | 免费商用授权 |
Local deploymentWeight size | 54.8GB | Not provided |
VRAM for weights | ≈ 54.8 GB (weights only, excludes KV cache) | Not provided |
Weights | Hugging Face | Hugging Face |
Modality supportText Input/Output | / | / |
Image Input/Output | / | / |
ResourcesPaper / report | Gemma 3 Technical Report | Gemma 4 Model Card |
DataLearner blog | Google开源第三代Gemma-3系列模型：支持多模态、最多128K输入，其中Gemma 3-27B在大模型匿名竞技场得分超过了Qwen2.5-Max | Google Gemma 4 正式开源：Apache 2.0 协议、手机端可运行、原生支持多模态和 Agent 工作流 |
