Gemma 4 Added to AI Gateway

Google's Gemma 4 26B (MoE) and 31B (Dense) are now available on Vercel AI Gateway. Both models are built on the same architecture as Gemini 3, and support function-calling, agentic workflows, structured JSON output, and system instructions. They also offer 256K context, 140+ languages, and native vision.

The two models serve different purposes.

• 26B (MoE): Activates only 3.8B parameters out of the total 26B during inference, optimized to reduce latency and increase tokens-per-second.
• 31B (Dense): Activates all parameters to focus on output quality, and is better suited as a base model for fine-tuning.

In the AI SDK, you can use these models by specifying the model name as google/gemma-4-31b-it or google/gemma-4-26b-a4b-it. AI Gateway consolidates model calls into a single API, and also provides usage/cost tracking, retry, failover, performance optimization, custom reporting, observability, BYOK, and intelligent provider routing.