Workers AI doesn't bill tokens — it bills neurons, Cloudflare's normalised unit of compute across every model type. The free allowance is a daily neuron budget, and each model has its own neuron cost per unit of work.
Why the unit matters: it makes model choice, not call count, the real lever. A small model might cost an order of magnitude fewer neurons per request than a large one, so the same daily budget is either a trickle of big-model calls or a flood of small-model ones. Text, embeddings and image models all draw from the same pool, which no token-based tier does — your embedding batch competes with your chat traffic.
Practical consequence: check the per-model neuron price before defaulting to the biggest model on the list, and remember the meter resets daily — unused budget doesn't roll.
Current model list and the neuron math per request: https://toolfreebie.com/cloudflare-workers-ai/
Top comments (0)