What Is Cloudflare Workers AI? Serverless AI, Free Neuron Allowance, Pricing, Models and Data

With Workers AI you run an AI model on Cloudflare's network without setting up your own server: models for text generation, images, speech recognition (Whisper) and text to speech (MeloTTS), for example. From your code you call a Workers binding or the REST API (with an OpenAI-compatible endpoint). The model pages say “LLM Playground requires no setup or authentication”; that's the way to try it without an account. The home page says “50+ models,” and the catalog says “70 models.”

Who it's for: developers who want to call models cheaply without managing their own infrastructure, and teams that use Cloudflare Workers.

!Even if you test Workers AI locally with wrangler dev, it accesses your Cloudflare account and usage is billed; the Limits page counts local inferences too.

Signing up asks only for an email and password; the documentation doesn't say whether a card or phone number is required. The account deletion steps aren't clear in the official documentation.

Workers AI isn't a single model but a catalog: the Meta Llama family, Mistral, Qwen, Whisper, MeloTTS, and Kimi, GLM and DeepSeek V4 models that require paid billing, for example. The catalog changes often; see the current list at developers.cloudflare.com/workers-ai/models.

!The @cf/meta/llama-3.1-8b-instruct model in the official REST and OpenAI-compatibility examples is on the pricing page but not in the catalog, and its model page returns a 404; its counterpart in the catalog has the -fp8 suffix. When you copy the example, confirm the model name in the catalog.

Check the number and names of models in the current catalog; the status of some models that are on the pricing page but don't appear in the catalog (such as kimi-k2.5) isn't clear. There's no official source on Turkish responses (Turkish isn't mentioned on the official pages).

!Despite the daily free allowance, some models don't run on the free plan (error code 5035, HTTP 403): Kimi K2.6, Kimi K2.7 Code, GLM 5.2, GLM 5.3, GLM 5.3 Flash, DeepSeek V4 Flash and DeepSeek V4 Pro require a paid billing method. For paid models, AI Gateway credit top-ups carry a 5% fee, and the request limit rises from 20 to 50 per minute.

Prices are in dollars; Turkish lira pricing and how tax is applied aren't stated. Separate pricing for the Batch API, student discounts and model speeds aren't in the official sources. On the pricing page, the Whisper and MeloTTS prices differ from the model pages by a rounding difference.

Official pricing page last checked on October 28, 2026.