Gemma 4 26B A4B IT
Weights availableGoogle open-weight Gemma 4 instruction-tuned mixture-of-experts multimodal model hosted by the Gemini API.
Access details: Gemini API
The Hugging Face repository. Download the weights and run them yourself or with a host.
gemma-4-26b-a4b-itUpdated Sep 23, 2026
Capabilities & limits
- Context window
- 256K
- Maximum output
- Not published
- Image input
- Supported
- Tool calling
- Supported
- Structured output
- Not published
Hosted price
USD / 1M tokens- Input
- $0.00
- Output
- $0.00
Full rate card 6 rates
All prices in USD.
Standard input
$0.00 per 1M tokens- Gemini API free tier
- paid tier not available
Standard output
$0.00 per 1M tokens- Gemini API free tier
- paid tier not available
Cached input
$0.00 per 1M tokens- Gemini API free tier
Cache storage
$0.00 per 1M tokens per hour- Gemini API free tier
Paid tier input
No longer sold per 1M tokens- paid tier
Paid tier output
No longer sold per 1M tokens- paid tier
Gemma 4 on the Gemini API is listed as free of charge on the free tier (content used to improve Google's products); every paid-tier price is listed as Not available. Tuning and Grounding with Google Search are not available. Self-hosting the open weights carries no provider token price.
Price history
Sep 23, 2026
Input $0.00 per million tokens
Sep 23, 2026
Output $0.00 per million tokens
Technical details
- Parameters
- 25.2B
- Active parameters
- 3.8B
- Architecture
- Mixture-of-Experts transformer: 128 experts, 8 active plus 1 shared; 30 layers; hybrid 1,024-token sliding-window and global attention; ~550M-parameter vision encoder
- Training / knowledge cutoff
- January 2025
- Input modalities
- Text, image
- Output modalities
- Text
License
Apache 2.0
From Google
Official announcement ↗Release & lifecycle
- Published
- Apr 2, 2026
- available
- As of Sep 13, 2026