Executive Summary
Vercel has announced immediate availability of Google's new Gemma 4 open models, including the 26B Mixture-of-Experts (MoE) and 31B Dense variants, on its AI Gateway platform. Built on the Gemini 3 architecture, these models provide developers with advanced capabilities such as function-calling, native vision, and a 256K context window. Integration through the Vercel AI Gateway offers a unified API for simplified access, usage tracking, and performance optimization.
Key Takeaways
* New Models Added: Google's Gemma 4 26B (MoE) and 31B (Dense) models are now accessible via Vercel.
* Two Model Variants:
* 26B (MoE): Optimized for low latency and speed, activating only 3.8B of its 26B parameters during inference.
* 31B (Dense): Designed for higher-quality output and is better suited as a base for fine-tuning.
* Advanced Capabilities: Both models support function-calling, agentic workflows, structured JSON output, native vision, and a large 256K context window.
* Platform Integration: Developers can call the models directly through the Vercel AI SDK using the identifiers `google/gemma-4-26b-a4b-it` and `google/gemma-4-31b-it`.
* Vercel AI Gateway Benefits: The platform provides a unified API, cost and usage tracking, observability, automatic retries, and failover management.
Strategic Importance
This integration provides Vercel developers with immediate access to state-of-the-art open models, solidifying Vercel's position as a competitive, full-stack AI application development platform.