← All models
Google · openrouter-31aaf493d17309eb51737c73
Gemma 4 26B A4B
Gemma 4 26B A4B IT is an instruction-tuned Mixture-of-Experts (MoE) model from Google DeepMind. Despite 25.2B total parameters, only 3.8B activate per token during inference — delivering near-31B quality at...
Available now
Blind win recordNot rated
Rated appearances0
Median latency8 s
Context window262K tokens
Average cost$0.0000
Capability profile
Where it performs well
- Agentic workTool use and multi-step workflows
- ReasoningMulti-step analysis and problem solving
- MultimodalUnderstanding beyond plain text
- Text generationInstruction following and response generation