Cloudflare Workers AI (Compute)

developers.cloudflare.com/workers-ai/platform/pricing/Inference APIcompute/cloudflare-workers-aiJSON compare report an error provider JSON

Free allowance: A daily allowance of 10,000 Neurons on both the Workers Free and the Workers Paid plan: 'Our free allocation allows anyone to use a total of 10,000 Neurons per day at no charge.' Limits reset daily at 00:00 UTC.
Verdict: caveatusable, with a named limit or an unprobed read path
Cost: free tier
Identity: emailno card, phone required
Binding limit: request-rateText generation is limited to 300 requests per minute, a default per task type that the limits page does not tie to one plan and that applies unless the model requires the Workers Paid plan. The models that require the Workers Paid plan...
Evidence: documented13 sources; a first-party page was read and the sentence quoted
Freshness: stale on 2026-11-29checked 2026-10-29 (live); stale on 2026-11-29

Free tier	A daily allowance of 10,000 Neurons on both the Workers Free and the Workers Paid plan: 'Our free allocation allows anyone to use a total of 10,000 Neurons per day at no charge.' Limits reset daily at 00:00 UTC. Neurons convert to tokens at a per-model rate, so the same 10,000 Neurons buys a different number of output tokens on each model: @cf/meta/llama-3.1-8b-instruct yields 133K output tokens a day at 75147 Neurons per 1M, @cf/qwen/qwen3-30b-a3b-fp8 yields 328K at 30475 per 1M and @cf/meta/llama-3.2-1b-instruct yields 548K at 18252 per 1M, a spread of about 4 across these three models. Eight models require a paid billing method rather than the Workers Free plan alone: kimi-k2.6, kimi-k2.7-code, glm-5.2, glm-5.3, glm-5.3-flash, deepseek-v4-flash-0731, deepseek-v4.1-flash-1017 and deepseek-v4-pro-0813. unit minutes; metered on wall-clock; recurring-daily-allowance
Price	none stated
Limits	request-rate (binding) Text generation is limited to 300 requests per minute, a default per task type that the limits page does not tie to one plan and that applies unless the model requires the Workers Paid plan. The models that require the Workers Paid plan are limited per account and per model to 20 requests per minute on standard Workers AI billing and 50 requests per minute with prepaid AI Gateway credits. Token volume is metered separately by the daily Neuron allowance on the pricing page. source
Identity	email a Cloudflare account created on the sign-up page with an email address and a password; Cloudflare then sends an email to verify the address. unverified: whether sign-up asks for a phone number or a card, which the create-account page does not name
Egress	not stated
Interface	rest inferred from the tooling the provider documents; not probed
Tooling	not stated
Composes with	the only free lane here with both a perpetual daily meter and 300 rpm: it pairs with a free Cloudflare Worker as the scheduler, so a nightly batch runs inside the free tier with no machine to keep awake.
Reliability	no reliability event recorded
Notes	All arithmetic is an estimate and not a vendor figure, but the conversion factor is checkable against the vendor's own two columns: USD 0.011 per 1,000 Neurons is USD 11 per 1M Neurons, and 34,868 Neurons per 1M output tokens for llama-3.1-8b-fp8-fast times USD 11 per 1M Neurons is USD 0.384, exactly the price printed beside it. The same check holds for qwen3-30b-a3b-fp8 (30,475 x USD 11 / 1M = USD 0.335). So Neurons are a linear repricing of GPU seconds and the free allowance has a fixed dollar value of 10,000 x USD 0.011 / 1,000 = USD 0.11 a day, or USD 3.30 a month, whatever model is chosen. The model choice therefore buys tokens, not dollars: 328K output tokens a day on qwen3-30b-a3b-fp8 is about 9.8M a month, and 548K on llama-3.2-1b is about 16.4M a month at visibly lower quality. Verdict is qualifies-with-caveat because no probe was run: the REST endpoint needs a real account id and an API token, and the keyless probe against a placeholder account id returned 404 from the origin, which establishes only that the route requires a credential.
Parsed numbers	request-rate 300 requests / minute

Evidence 13

date	source	what it proved
2026-10-22	developers.cloudflare.com/workers-ai/platform/pricing/	'Our free allocation allows anyone to use a total of 10,000 Neurons per day at no charge' and '$0.011 per 1,000 Neurons', quoted
2026-10-22	developers.cloudflare.com/workers-ai/platform/pricing/	@cf/qwen/qwen3-30b-a3b-fp8 at 4625 neurons per M input and 30475 per M output against $0.051 and $0.335 per M tokens; @cf/meta/llama-3.1-8b-instruct at 25608 and 75147 neurons against $0.282 and $0.827, quoted
2026-10-22	developers.cloudflare.com/workers-ai/platform/pricing/	'Some models require a paid billing method' listing the eight excluded models, quoted
2026-10-22	developers.cloudflare.com/workers-ai/platform/limits/	Text Generation 300 requests per minute; paid-plan models 20 rpm on standard Workers AI billing, quoted
2026-10-22	api.cloudflare.com/client/v4/accounts/test/ai/run/@cf/meta/l	not published
2026-10-22	developers.cloudflare.com/workers/platform/limits/	quoted table: CPU time per HTTP request, Workers Free 10 ms, Workers Paid 5 min default 30 seconds; CPU time per Cron Trigger, Free 10 ms
2026-10-22	developers.cloudflare.com/workers/platform/limits/	quoted: 'The average Worker uses approximately 2.2 ms per request. Heavier workloads that handle authentication, server-side rendering, or parse large payloads typically use 10-20 ms'
2026-10-22	developers.cloudflare.com/support/troubleshooting/http-statu	quoted: Error 1102 means the Worker 'has exceeded its CPU time limit or memory limit'; resolution requires the Paid plan to raise the limit
2026-10-22	developers.cloudflare.com/workers/platform/limits/	quoted: 'Each isolate has some built-in flexibility to allow for cases where your Worker infrequently runs over the configured limit. If your Worker starts hitting the limit consistently, its execution will be terminated'
2026-10-22	community.cloudflare.com/t/what-happens-when-the-cpu-time-is	community answer on the burst behaviour: small bursts over 10 ms are allowed, and a worker that goes beyond is shut down [date precision: year precision] [re-read 2026-10-22]
2026-10-28	developers.cloudflare.com/workers-ai/platform/pricing/	the Workers Free plan carries 10,000 Neurons per day, as does Workers Paid; the table prints no price on Workers Free
2026-10-28	developers.cloudflare.com/workers-ai/platform/limits/	Text Generation is limited to 300 requests per minute unless the model requires the Workers Paid plan; the paid-model table gives 20 and 50 requests per minute
2026-10-28	developers.cloudflare.com/fundamentals/account/create-accoun	account creation asks for an email address and a password, then emails a verification message

More fields 1

{
 "facets": {
  "class": "perpetual-free-allowance",
  "free_tier_kind": "recurring-daily-allowance",
  "shape": {
   "vcpu": null,
   "ram_gib": null,
   "gpu_class": null,
   "gpu_count": null
  }
 }
}

Unknown values 6

$.egress.metered egress metering was not published
$.egress.terms a source was read and is silent on this field
$.per_object_cap.stated the field does not apply to this kind of offer
$.reliability_note nothing of this kind was observed
$.retention.stated the field does not apply to this kind of offer
$.tooling a source was read and is silent on this field