Cloudflare's New Open-Weight Model Fits Inside the Free Daily Allowance

An extraction model priced so low that the free tier stops being a trial and starts being a workload, for exactly thirteen more days.

Share
Cloudflare's New Open-Weight Model Fits Inside the Free Daily Allowance

Cloudflare released an updated Clairic and a new Clairic-flash as open-weight extraction models on 18 November 2026, both hosted on Workers AI. Clairic-flash is a 9-billion-parameter model at $0.10 per million input tokens, which is 9,091 Neurons; Clairic is the 27-billion-parameter version at $0.24 per million, or 21,818 Neurons. Both charge on input tokens only, with Cloudflare absorbing output generation, and Clairic is on a promotional 58% discount at $0.10 until 1 December 2026.1

The reason this matters more than a model release usually does is the free allowance underneath it. Workers AI gives every plan, including Workers Free, 10,000 Neurons per day, and these are among the few models where that allowance buys a real volume of work rather than a test.2

How far do 10,000 Neurons go?

ModelPrice per M input tokensNeurons per MFree daily tokens
Clairic-flash (9B)$0.1009,091~1.1 million
Clairic (27B), promo to 1 Dec$0.1009,091~1.1 million
Clairic (27B), from 1 Dec$0.24021,818~458,000

Prices measured on 27 October 2026. Vendor plan terms, allowances and prices change without notice, so treat every figure on this page as a dated snapshot rather than a permanent fact. Where a number here differs from the vendor's own pricing page, the vendor's page is correct and the figure above is simply older than it.

Roughly a million tokens of input a day, at no charge, on a model purpose-built for pulling structured fields out of documents. A receipt is a few hundred tokens. An invoice might be two thousand. A million tokens a day is thousands of documents daily, sustained, on a free plan.

That is an unusual position for a free tier. Most free allowances are sized to let you evaluate a product, and this one is sized to run a small back-office job. The limits reset daily at 00:00 UTC.

PlanIncludedBeyond the allowance
Workers Free10,000 Neurons per dayNot available, upgrade required
Workers Paid10,000 Neurons per day$0.011 per 1,000 Neurons

What changes on 1 December?

Clairic-flash stays at $0.10. The larger Clairic does not: the promotional discount ends on 1 December 2026 and the price returns to $0.24 per million input tokens, which is 21,818 Neurons.

The practical consequence on a free plan is that the daily ceiling for the large model more than halves, from roughly 1.1 million input tokens to about 458,000. A workflow tuned to 27B Clairic in the next thirteen days will need either a cheaper model, a shorter day, or a paid plan in December.

Cloudflare is direct about the intended path: move cost-sensitive workloads to the cheaper 9B variant. That is also the honest reading of the pricing, because the two models do not price at a premium by accident.

What do the models actually do?

These are extraction models, not chat models. You send a document, a schema of fields and receive typed answers with confidence scores, character spans and probabilities, which is what makes the output usable by software rather than by a reader.

Cloudflare reports a redesigned architecture using parallel slot autoregressive generation, where each field generates its own tokens and stops at its own end marker while all fields advance together in the serving batch. Its internal evals show Clairic at 335 of 336 on fresh text fields and 1,263 of 1,263 on tracked customer assertions, against 1,033 for the preview model. Those are self-reported and there are no public extraction benchmarks, which the post says plainly.

The latency numbers are the more useful claim because they are harder to dress up. Clairic-flash runs a median of 155 ms on fresh text fields and 305 ms on fresh images, with p95 at 328 ms and 605 ms. The larger model is roughly twice as slow and still under a second at the median across every evaluated set.

Documents can now include up to four images or PDF pages in one request, with line items returned as rows that keep description, quantity and price together. That row structure is the difference between an extraction API and a text generator with a good prompt.

How does this compare with other free tiers?

According to Cloudflare, the goal is to help teams augment expensive LLM usage with smaller, cheaper, purpose-built models. Pricing at input-only, with output absorbed, pushes that argument further than most.

Against the rest of the free-tier landscape this is the most generous per-day allowance we have found for a task-specific model. When we checked the Workers AI pricing page alongside the announcement, the free allocation was the same 10,000 Neurons it has been, so nothing about the floor changed. What changed is that a capable extraction model now fits under it.

The verdict is that Clairic-flash is genuinely usable on a free plan today, and that the 27B model is worth evaluating in the next thirteen days precisely because it costs the same until 1 December.

The limitations deserve naming. The promotional price expires and the free daily ceiling for the large model more than halves. The benchmarks are self-reported with no public extraction evaluation to check them against. A request carries at most four images or PDF pages. Exceeding the daily allowance stops the work rather than billing it, so a job begun late in the UTC day simply fails. And some Workers AI models now require a paid billing method, though these two do not.

Best for document and receipt extraction on a small budget. Not ideal for anything needing more than a million tokens a day without a paid plan.

Frequently asked questions

Can I use Clairic on a free Workers plan?
Yes. Workers Free includes 10,000 Neurons per day, and Clairic-flash fits roughly a million input tokens into that.

How much does Clairic cost?
$0.10 per million input tokens until 1 December 2026, then $0.24 per million.

How much does Clairic-flash cost?
$0.10 per million input tokens, unchanged, which is 9,091 Neurons.

Am I charged for output tokens?
No. Both models charge on input only and Cloudflare absorbs output generation.

What happens if I exceed the daily allowance?
Operations fail with an error. Workers Free cannot continue until the 00:00 UTC reset.

Are the weights available?
Yes, both are open-weight on Hugging Face as well as hosted on Workers AI.

The model sizes, endpoints, prices, the 58% promotional discount to 1 December 2026, the input-only billing, the architecture, the internal evals, the latency percentiles, the four-document limit and the open-weight release are from Cloudflare's announcement of 18 November 2026. The 10,000 Neuron daily free allocation, the $0.011 per 1,000 Neurons paid rate, the 00:00 UTC reset, the Clairic Neuron-equivalents and the note that some models require a paid billing method are from Cloudflare's Workers AI pricing documentation, read on 19 November 2026.

Sources

  1. Cloudflare Blog: Open Clairic: blog.cloudflare.com/open-clairic. Used for: Clairic 27B $0.24 per M input, promotional $0.10 until 1 December 2026; Clairic-flash 9B $0.10.
  2. Cloudflare Docs: Workers AI pricing: developers.cloudflare.com/workers-ai/platform/pricing. Used for: Workers Free 10,000 Neurons per day; paid $0.011 per 1,000 Neurons; resets 00:00 UTC.

All sources are the vendor's own plan or documentation pages, checked on 27 October 2026. Plan terms change without notice, so confirm current limits before you rely on them.