Budget Your Free Daily Neurons on Cloudflare Workers AI

Why this works

Cloudflare's pricing page says Workers AI is included in both the free and paid Workers plans, with a free allocation of 10,000 Neurons per day, and that a day's limit is reset at 00:00 UTC. Neurons are a unit of GPU work that differs per model, so the first skill is reading the price table for the model you pick.

Step by step

1. Read the model's price in NeuronsOn the pricing page, each model lists Neurons per million input and output tokens. For example, the small @cf/meta/llama-3.2-1b-instruct lists 2,457 Neurons per million input tokens and 18,252 per million output tokens.
2. Do the sum for your workloadEstimate tokens per request and requests per day. Roughly: 10,000 Neurons covers about 4 million input tokens on that small model, but only about half a million output tokens, because output costs more Neurons.
3. Pick the cheapest model that worksTest two or three models on your real prompts and compare answer quality against the Neuron cost, not the headline price alone.
4. Cap your usageAdd a counter in your Worker that stops calling the model at a daily threshold you set, and falls back to a message to the user.
5. Watch the dashboardCheck the Workers AI usage view after the first week, then once a month.

Copy-paste prompt

For a Cloudflare Worker that calls a Workers AI model, outline: (1) the daily request budget from my 10,000-Neuron allocation given the model's Neuron price in my table, (2) a daily counter that stops calls at 90% of budget, (3) the message users see when the budget is reached. Show the calculation.

MODEL PRICE TABLE:
<paste>
AVERAGE TOKENS PER REQUEST:
<number>

Check your result

• The calculation uses the table for your exact model
• A daily stop is in place before you publish
• You know when the UTC day resets

Pitfalls to avoid

• Output tokens are the expensive part; ask for short answers.
• Some models are priced differently; read the entry for your model, not a neighbouring one.