What models are supported by the Cloudflare Workers AI skill?

It supports the latest 2025 models including Llama 4 Scout, GPT-OSS 120B, Mistral 3.1 (Vision/Tools), Gemma 3, and specialized models like Flux for images and Deepgram for audio.

Why is streaming recommended for text generation?

Streaming prevents Worker memory buffering issues, avoids execution timeouts, and provides a much faster time-to-first-token (TTFT) for better user experiences.

What are the 2025 breaking changes for embeddings?

BGE models now support a 'pooling: cls' parameter which is 2x faster but not backwards compatible with the previous 'mean' default, requiring a metadata migration for existing vector indexes.

How does the AI Gateway integration help?

The AI Gateway provides a unified interface for caching requests to save costs, detailed logging for debugging, and analytics to track neuron usage across different models.

Cloudflare Workers AI

Name: Cloudflare Workers AI
Author: jezweb

byjezweb

•

Data Science & ML

Enables high-performance AI model inference on Cloudflare’s global GPU network for text, image, and embedding generation.

About

This skill equips Claude with the expertise to implement, optimize, and troubleshoot Cloudflare Workers AI integrations. It provides deep technical guidance for the 2025 model lineup—including Llama 4 Scout, GPT-OSS, and Gemma 3—while ensuring developers follow best practices for streaming, vision tasks, and audio processing. From building high-speed RAG systems with BGE embeddings to managing costs via AI Gateway, this skill handles the complexities of serverless AI, including critical 2025 breaking changes like BGE pooling parameters and max_token default shifts.

Key Features

Multi-modal capabilities for Flux image generation and Deepgram audio processing.
91 GitHub stars
AI Gateway configuration for request caching, logging, and granular cost tracking.
Support for 2025 models including Llama 4, GPT-OSS 120B, and Mistral 3.1 Vision.
Optimized RAG implementation using high-speed BGE and Gemma embeddings.
Native streaming integration to eliminate worker timeouts and improve TTFT.

Use Cases

Building low-latency, serverless RAG applications with Vectorize integration.
Implementing real-time AI chat interfaces with event-stream support.
Automating high-fidelity image generation and audio transcription pipelines.

About

Key Features

Multi-modal capabilities for Flux image generation and Deepgram audio processing.
91 GitHub stars
AI Gateway configuration for request caching, logging, and granular cost tracking.
Support for 2025 models including Llama 4, GPT-OSS 120B, and Mistral 3.1 Vision.
Optimized RAG implementation using high-speed BGE and Gemma embeddings.
Native streaming integration to eliminate worker timeouts and improve TTFT.

Use Cases

Building low-latency, serverless RAG applications with Vectorize integration.
Implementing real-time AI chat interfaces with event-stream support.
Automating high-fidelity image generation and audio transcription pipelines.