Cloudflare Workers AI Neurons Cost Calculator
✓ Last verified: 2026-09-01· Neuron rates shown are Cloudflare's published per-model rates, not live quotes· Reference estimate — report change → Cloudflare Workers AI bills every model — text generation, image generation, embeddings, whatever you run — in one normalized unit called a Neuron, priced at $0.011 per 1,000 Neurons. Every account gets 10,000 Neurons/day free, shared across your entire account regardless of which models you call, resetting daily at 00:00 UTC. That's a genuinely different billing shape from a single model's $/million-token rate: a Llama-3.2-1B chat feature, a Flux-1-schnell image generator and a BGE embeddings pipeline all draw from the same daily allowance, so the mix of workloads — not just the volume — determines when you start paying. This calculator combines text, image and embedding usage into one Neuron total, applies the shared free tier, and prices the overage. For raw hyperscaler token pricing see the context window cost calculator; for edge-vs-cloud tradeoffs see the edge/on-device AI vs cloud calculator.
—Neurons used per day
—Neurons billed per day (over free 10,000)
—billed cost per day
Biggest driver of your Neuron bill
—
Neuron breakdown by workload type
How your daily 10,000-Neuron free allowance is split across the three workload types at the settings above.
| Workload | Neurons/day | % of total |
|---|
Monthly cost by text requests/day
Image and embedding usage held at the values above, swept across text requests per day, showing where the shared 10,000/day free allowance runs out. The highlighted row is closest to your current input.
| Text req/day | Total Neurons/day | Overage Neurons/day | Monthly cost |
|---|
How this connects to other tools
This calculator prices one specific mechanic: Cloudflare Workers AI's Neuron unit, which normalizes text generation, image generation and embeddings into one number and bills them against a single shared daily free allowance. That's different from the Cloudflare Workers cost calculator, which prices Workers' serverless compute (CPU time and requests) rather than Workers AI's model inference. It's also a different question from raw GPU rental — see the GPU inference cost calculator for self-hosted inference economics, and the edge/on-device AI vs cloud calculator for when running inference at the edge beats a centralized cloud API at all. For the image and embedding halves of this calculator priced on their own without the shared Neuron pool, see the image generation cost calculator and the embeddings cost calculator.
How this calculator works
The Cloudflare Workers AI Neurons Cost Calculator converts three different workload types into one shared unit. Text: at the defaults (Llama 3.1 70B, 300 requests/day, 800 input + 300 output tokens/request), each request costs 800/1,000,000 × 26,668 + 300/1,000,000 × 204,805 = 21.33 + 61.44 = 82.78 Neurons, so 300 requests/day burns roughly 24,833 Neurons/day. Image: Flux-1-schnell at 1 tile (512×512) and 4 steps costs 1 × 4.80 + 4 × 9.60 = 43.20 Neurons/image, so 20 images/day adds 864 Neurons/day. Embeddings: BGE-small at 100,000 input tokens/day costs 100,000/1,000,000 × 1,841 = 184.1 Neurons/day. Total usage is roughly 25,881 Neurons/day.
Against the 10,000 Neurons/day free allowance (shared across all three workload types, reset daily at 00:00 UTC), that leaves about 15,881 overage Neurons/day, billed at $0.011 per 1,000 Neurons: 15.881 × $0.011 ≈ $0.175/day, or roughly $5.24/month at 30 active days. Text generation is the dominant driver at these settings — about 96% of daily Neurons — so the model choice for text (70B vs 1B) matters far more to the bill than the image or embedding volume. Because the free tier is one shared account-wide pool rather than one allowance per model, adding image or embedding calls to an already text-heavy app pushes you into paid Neurons faster than either workload would alone.
Frequently asked questions
What is a Cloudflare Workers AI Neuron and how is it calculated?
A Neuron is Cloudflare's normalized unit for measuring the GPU compute an AI request consumes on Workers AI, billed at $0.011 per 1,000 Neurons. Each model has its own Neuron rate per unit of work: Llama-3.2-1B costs 2,457 Neurons per million input tokens and 18,252 per million output tokens; the larger Llama-3.1-70B costs 26,668 per million input tokens and 204,805 per million output tokens; Flux-1-schnell image generation costs 4.80 Neurons per 512×512 tile plus 9.60 Neurons per step; BGE-small embeddings cost 1,841 Neurons per million input tokens and BGE-large costs 18,582. Because every model type converts into the same Neuron unit, a text-heavy workload and an image-heavy workload can be compared and billed on one line.
How much of Cloudflare Workers AI is free?
Every account gets 10,000 Neurons per day free, on both the Workers Free and Workers Paid plans, resetting daily at 00:00 UTC. That allowance is a single shared pool across your entire account — it doesn't matter whether the Neurons come from text generation, image generation or embeddings, they all draw from the same daily 10,000. Once you exceed it on a given day, the overage is billed at $0.011 per 1,000 Neurons, which requires being on the Workers Paid plan.
Why does mixing model types in one app matter for the Neuron bill?
Because the 10,000 Neurons/day free allowance is shared across every model you call, not allocated per model. An app that calls a small text model, generates a few images and embeds documents all draws down the same daily pool, so the order and mix of calls determines how fast you burn through the free tier. Model choice matters enormously too: Llama-3.1-70B burns roughly 11x more Neurons per input token and 11x more per output token than Llama-3.2-1B, so switching to the larger model on a busy day can turn a $0 bill into a paid one even with identical traffic volume.
How does Workers AI Neuron pricing compare to raw token pricing on OpenAI or Anthropic?
Hyperscaler LLM APIs like OpenAI and Anthropic bill directly in dollars per million tokens, with a different input and output rate for every model and no unified daily free allowance shared across model types — see our context window cost calculator and 387-model price comparison for that shape. Workers AI instead converts every model's compute cost into one abstract unit (the Neuron) billed at a single flat rate of $0.011 per 1,000 Neurons, with a 10,000 Neuron/day account-wide free tier that applies no matter which models you mix. The tradeoff is that you're running smaller, Cloudflare-hosted open models (Llama, Flux, BGE) at the edge rather than frontier closed models, in exchange for one predictable per-unit rate and free daily headroom.