Cloudflare Workers AI

Offering overview

Inference that runs wherever the request lands is the idea behind Workers AI. Open models from Meta Llama, DeepSeek, Mistral, Google Gemma, Qwen, OpenAI gpt-oss, IBM Granite, GLM, Nemotron and Kimi, along with Whisper, FLUX, BGE embeddings and translation and vision models, are hosted on GPUs in Cloudflare's datacentres and invoked from Workers code or a REST API. It pairs with Vectorize for vector search, AI Gateway for caching, rate limiting and observability, R2 for storage and Durable Objects for agent state, giving a complete edge stack for RAG and agent applications. There are no instances to size or scale; capacity is elastic and included in both Free and Paid Workers plans. Front-end and full-stack developers who already build on Workers get the most out of it. Usage is metered in neurons at USD 0.011 per 1,000 after 10,000 free neurons per day, which translates to about USD 0.017 per 1M input tokens for Granite 4.0 H Micro and higher rates for larger LLMs.

Technical specifications

Containers

Unknown

Kubernetes

Unknown

Root access

Unknown

API access

Unknown

Autoscaling

Unknown

Persistent processes

Unknown

AI capabilities

Model serving

Llama, DeepSeek, Mistral, Gemma, Qwen, OpenAI gpt-oss, IBM Granite, GLM, NVIDIA Nemotron, Kimi, SEA-LION, Whisper, FLUX, BGE, Qwen3 Embedding, MeloTTS

AI frameworks

Not stated

Inference endpoints

Unknown

Fine-tuning

Unknown

Vector database

Unknown

Model catalog

Unknown

Monitoring

Not stated

Evaluation

Not stated

Security & compliance

Geography

Provider headquarters are separate from offering infrastructure and availability.

Available plans

Pricing can change. Promotional prices are shown separately from standard prices.