Gemma 4
Gemma 4 is an open-weight multimodal family with native thinking and function calling, released under Apache 2.0. The initial E2B and E4B models target constrained devices, while 26B A4B uses sparse experts and 31B is dense. All support vision and video frames; audio is available in the smaller models. Context lengths vary by size, and the independently released encoder-free 12B update has its own page.
2026-04-02
E2B: 2.3B effective / 5.1B total; E4B: 4.5B effective / 8B total; 26B A4B: 25.2B total / 3.8B active; 31B: 30.7B
Decoder-only Transformer family with local/global attention; dense E2B, E4B and 31B; sparse mixture of experts for 26B A4B (8 of 128 experts plus one shared)
Apache-2.0
Specifications
- Parameters
- E2B: 2.3B effective / 5.1B total; E4B: 4.5B effective / 8B total; 26B A4B: 25.2B total / 3.8B active; 31B: 30.7B
- Architecture
- Decoder-only Transformer family with local/global attention; dense E2B, E4B and 31B; sparse mixture of experts for 26B A4B (8 of 128 experts plus one shared)
- License
- Apache-2.0
- Context Window
- 262,144 tokens
- Training Data Cutoff
- 2025-01
- Type
- multimodal
- Modalities
- textimageaudiovideo
Benchmark Scores
Advanced Specifications
- Model Family
- Gemma
- API Access
- Not Available
- Chat Interface
- Not Available
- Multilingual Support
- Yes
- Variants
- gemma-4-E2B (128K context)gemma-4-E4B (128K context)gemma-4-26B-A4B (256K context)gemma-4-31B (256K context)Pre-trained and instruction-tuned variants
- Hardware Support
- GPUTPUCPU
Capabilities & Limitations
- Capabilities
- reasoningcodevisionvideo framesaudio understandingfunction callingmultilingual
- Known Limitations
- Audio input is limited to 30 seconds and is supported only by E2B and E4B among the original sizes.Memory includes all weights and grows with context; active parameter count does not equal required weight memory.Generated content can contain factual errors and bias.
- Notable Use Cases
- local assistantsmultimodal analysistool-based on-device applications
- Function Calling Support
- Yes
- Tool Use Support
- Yes