Gemma-4-26B-A4B
- Name in the API
google/gemma-4-26B-A4B-it- Base model
- google/gemma-4-26B-A4B-it
- Weights we run
- RedHatAI/gemma-4-26B-A4B-it-FP8-dynamic (FP8 quantization)
- Type
- Chat
- Input → output
- text, image → text (audio input is switched off)
- Context length
- 262,144 tokens
- Availability
- Dedicated
- Hardware
- One GPU with at least 90 GB (NVIDIA H100 94 GB or RTX PRO 6000 96 GB)
- Thinking
- Default: off.
- Turn on: any
reasoning_effort other than none (even minimal), or chat_template_kwargs: {"enable_thinking": true}.
- The level does not change the amount of thinking: in our tests all levels from
minimal to max
thought about the same amount.
- Numeric effort: no.
- Tool calling
- Yes
- Example
- Chat request, question about an image. Set
model to the name above.