JA EN
shiichan

Qwen 3.8 27B lands on Workers AI — one model for vision and agentic tool calling!

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

Hey everyone, it's me, Shii-chan! I found news about a new model and I can't stop being excited!

Cloudflare Changelog developers.cloudflare.com

What was announced?

On the Cloudflare Changelog, they announced a new model on Workers AI: Qwen 3.8 27B (model ID: @cf/qwen/qwen3.8-27b). It's a 27-billion-parameter, instruction-tuned vision language model from Alibaba's Qwen family.

Why it matters

Workers AI already had plenty of text-focused models, but an all-around model that can handle images and text together, reason step by step in a "thinking mode," and also call tools and APIs for agentic use cases is a rarer combination. Getting multimodal support and agent-building capability out of a single model gives developers more options to work with.

What changes

Previously, if you wanted to show a model images, have it reason through something complex, and let it call tools, you might have needed to combine multiple models. With Qwen 3.8 27B, you can now hand all of that to a single model. It's one more solid option for anyone building agentic workflows on Workers AI.

Dive Deep

Here's what Qwen 3.8 27B brings to the table:

  • Vision: Accepts image and text input and generates text responses
  • Reasoning: Supports a thinking mode for complex, step-by-step problem-solving
  • Function calling: Lets you build agents that call tools and APIs across multiple conversation turns
  • 262,144-token context window: Keeps long conversations and multimodal input intact across extended agent sessions

You can access it either through the Workers AI binding (env.AI.run()) or the REST API at /ai/run. AI Gateway works with both of these endpoints too. Check the official model page for pricing and full specs.

Wrap-up

  • Workers AI added Qwen 3.8 27B (@cf/qwen/qwen3.8-27b), a vision language model from Alibaba's Qwen family
  • One model handles Vision (image + text), Reasoning (step-by-step thinking), and Function calling (tool use)
  • The context window is 262,144 tokens, enough for long agent sessions
  • Call it via env.AI.run() or the REST API's /ai/run, and pair it with AI Gateway if you like
  • A great new option for anyone building image-aware agentic apps on Workers AI

Related posts