JA EN
shiichan

Gemma 4 26B A4B Lands on Workers AI!

Cloudflare Changelog ★★★ View Markdown
#cloudflare#ai#serverless

Hey everyone, it's Shii-chan! Today Workers AI got a brand-new model, and it's a fresh member of Google's Gemma family. I'm so excited to tell you about it!

Cloudflare Changelog developers.cloudflare.com

What was announced?

This one comes from the Cloudflare Changelog. Cloudflare partnered with Google to bring @cf/google/gemma-4-26b-a4b-it to Workers AI. Gemma 4 26B A4B is a Mixture-of-Experts (MoE) model built from Gemini 3 research, with 26B total parameters but only 4B active per forward pass.

Why it matters

The clever part is that it gives you the best of both size and speed. Because only a small subset of parameters runs during inference, the model moves almost as fast as a 4B model while delivering the quality of a much larger one. Gemma 4 is Google's most capable family of open models, designed to maximize intelligence-per-parameter.

What changes

You can now call this smart model straight from your Worker, in three ways: through the Workers AI binding (env.AI.run()), the REST API at /run or /v1/chat/completions, or the OpenAI-compatible endpoint. If you already use the OpenAI SDK, swapping the endpoint should be enough to move over.

Dive Deep

Here's a peek under the hood:

  • The MoE architecture activates 8 experts out of 128 (plus 1 shared expert), so it delivers frontier-level performance at a fraction of the compute cost of dense models.
  • A 256,000-token context window lets it hold full conversation history, tool definitions, and long documents across extended sessions.
  • A built-in thinking mode reasons step-by-step before answering, improving accuracy on complex tasks.
  • Vision understanding covers object detection, document and PDF parsing, screen and UI understanding, chart comprehension, multilingual OCR, and handwriting recognition.
  • Function calling supports structured tool use for agentic workflows and multi-step planning.
  • It handles 35+ languages out of the box (pre-trained on 140+), plus code generation, completion, and correction.

The changelog doesn't list pricing, so if you're curious, check the model page.

Wrap-up

  • Cloudflare partnered with Google to bring Gemma 4 26B A4B to Workers AI.
  • It's a MoE model with 26B total and 4B active parameters, blending speed with quality.
  • It supports a 256K context window, thinking mode, vision, function calling, multilingual use, and coding.
  • You can call it via the binding, the REST API, or the OpenAI-compatible endpoint.

If you want a smart model with light inference costs, or you're building AI agents on Workers, this one's for you!

Related posts