Cloudflare has introduced Clairic, a specialised artificial intelligence model designed to extract, calculate, estimate and forecast values from unstructured information. Announced on October 19, 2026, Clairic is available in preview through Cloudflare Workers AI. Instead of generating paragraphs of text like a conventional language model, it returns structured values such as invoice totals, dates, durations, estimates and future projections. The company is positioning it as a faster and lower-cost option for repetitive data-processing tasks that developers often assign to larger AI models.
The launch introduces a new category that Cloudflare calls System One extraction models. Clairic processes input in a single forward pass and returns values according to a developer-defined schema, along with confidence information and uncertainty estimates where appropriate. Its pricing is $0.10 per million input tokens, with no output-token charges. Cloudflare plans to release the model’s weights openly in the future, but the initial announcement provides hosted preview access rather than an open-weight release. The company also acknowledges limitations in multi-step arithmetic and forecasting, meaning developers will need to validate results before relying on them in critical workflows.
Key Takeaways
- New AI model: Cloudflare launched Clairic on October 19, 2026, through Workers AI in preview.
- Core functions: It can extract, compute, estimate and forecast values from unstructured inputs.
- Structured responses: Outputs include typed fields such as numbers, dates, durations, lists and forecasts.
- Pricing: The model costs $0.10 per million input tokens, with no output-token charges.
- Developer access: Applications can call Clairic through Cloudflare Workers AI or its REST API.
- Open weights planned: Cloudflare intends to release the model weights in the future, but has not made them openly available with this preview launch.
- Important limitation: Typed outputs are not necessarily correct, and complex calculations or long-range forecasts may require separate tools and human review.
What Is Cloudflare Clairic?
Clairic is an extraction model built for applications that need specific values from documents, messages, reports and other data sources.
Traditional language models can already perform these tasks. A developer can give an AI system an invoice and ask it to return the invoice number, total amount, vendor and payment deadline. However, the model may return incorrectly formatted data, misread a number or generate a value that does not appear in the original document.
Developers often add validation rules, parsing tools, retries and manual checks to reduce these errors. These additional steps can increase complexity, processing time and cost.
Clairic takes a different approach. Developers specify the fields they want, their expected data types and optional instructions. The model then returns the corresponding values rather than generating a conventional written answer.
For example, an application could process a customer email and request the order number, delivery date, total amount paid and number of days since delivery. Clairic can return these as separate, typed fields, making them easier for software to use directly.
Cloudflare says the model is designed to handle such tasks in one pass without generating output tokens. Its objective is to make repetitive extraction work more efficient than using a general-purpose language model for every field.
How the System One Extraction Model Works
Cloudflare describes Clairic as a System One model, a term inspired by the distinction between fast, intuitive processing and slower, deliberate reasoning.
In Cloudflare’s model framework, extraction systems focus on producing specific values quickly. Decision models select among predefined options, while general-purpose language models handle open-ended generation and more complex reasoning.
The company’s earlier decision models, Clef and Clef Flash, are designed to choose among options supplied by developers. Clairic expands the approach by returning values that cannot necessarily be listed in advance, such as a bill’s total amount or an estimated repair cost.
| Model category | Primary function | Example task |
|---|---|---|
| Extraction model | Extract, compute, estimate or forecast values | Read an invoice total and calculate the due date |
| Decision model | Choose among predefined options | Classify an invoice for approval or review |
| Generative language model | Produce open-ended text or reasoning | Explain a billing discrepancy to a customer |
The three approaches can be combined in an automated workflow. Clairic can first extract relevant information, a decision model can determine what action is appropriate, and a language model can generate an explanation or communicate with the user.
This division of work allows developers to use specialised models for specific tasks instead of relying on a single general-purpose system to perform every step.
Four Things Clairic Can Do With Unstructured Data
1. Extract values from documents
Clairic can identify information explicitly present in a document, including names, order numbers, dates, descriptions and amounts.
It can also return the location of extracted text within the original input through character offsets. This helps developers trace a value back to its source instead of receiving an answer without a clear reference.
For example, an invoice-processing application could extract the supplier name, invoice number, issue date and amount due from a document. The returned fields can then be passed into accounting software or checked against a purchase order.
The model also supports typed outputs, which helps applications distinguish between a date, a monetary amount and a text field.
2. Compute values from supplied information
The model can derive certain values using facts in the input. For instance, it may calculate a payment deadline from an invoice date and payment terms, or work out how many days a shipment is late.
This is different from extracting a value that is already written in the document. Computation requires the model to combine information supplied in the input.
Cloudflare says simple calculations, including one- and two-step arithmetic and date calculations, appear reliable in its early testing. Longer chains of dependent calculations and aggregation across many rows remain more difficult.
For applications that require precise arithmetic, the company recommends extracting the relevant inputs with Clairic and performing the calculations in conventional code.
3. Estimate values that are not explicitly stated
Some applications need estimates rather than exact information from a document.
A repair request, for example, may describe damage without specifying the final repair cost. A model could estimate the amount using comparable examples or descriptive information supplied in the input.
Clairic can return a distribution of possible values within boundaries specified by the developer. This includes a median and an 80% interval, giving applications more information about uncertainty than a single number alone.
However, estimates are only as useful as the available evidence. Cloudflare says Clairic performs best when the input includes comparable values, historical data or descriptive details. Asking for an estimate without meaningful evidence may produce a broad or unreliable range.
4. Forecast future values
Clairic can also forecast values from historical time-series information. A developer could provide previous sales figures, traffic measurements or other numerical history and request estimates for future periods.
The model returns a forecast for each requested step, with an 80% uncertainty interval. Cloudflare’s documentation supports forecasts of up to 64 steps, depending on the requested configuration.
This capability could help developers build lightweight forecasting features without creating a separate machine-learning pipeline for every metric.
However, Cloudflare explicitly warns that Clairic’s forecasting remains basic. Forecasts may flatten toward an average over longer horizons and cannot anticipate events that are not reflected in the input data. The company does not position the model as a replacement for specialised time-series models trained on relevant datasets.
Why Cloudflare Is Targeting Repetitive AI Workloads
Many business applications use AI for narrow tasks such as reading invoices, processing customer support messages, extracting shipping information and organising records.
These tasks may not require an elaborate explanation or an extended reasoning process. They often require a small number of specific values that software can use immediately.
Using a large language model for every field can introduce unnecessary overhead. Developers may need to pay for generated output tokens, parse the response and validate its structure before using the result.
Clairic’s approach aims to reduce this burden by returning typed values directly. Cloudflare says developers can request multiple fields in a single operation, with fields processed in parallel when they fit within the model’s processing constraints.
Potential applications include:
- Invoice and receipt processing.
- Customer support and returns workflows.
- Shipping and delivery tracking.
- Document classification pipelines that require extracted values.
- Lightweight estimates based on historical or descriptive data.
- Forecasting features for non-critical business metrics.
The practical benefit will depend on how well the model performs on a particular dataset, how much validation is required and whether its cost advantage persists in real applications.
Clairic Pricing and Availability
Clairic is available in preview through Cloudflare Workers AI under the model identifier @cf/cloudflare/clairic. Developers can call it through a Workers AI binding or the REST API.
| Detail | Information |
|---|---|
| Model | Clairic |
| Provider | Cloudflare |
| Availability | Preview |
| Platform | Cloudflare Workers AI |
| Input pricing | $0.10 per million input tokens |
| Output-token charge | None |
| Input types | Text and up to four accompanying images |
| Output types | Typed values, lists, estimates and forecasts |
| Open-weight release | Planned for the future; not yet released with the preview |
The absence of an output-token charge follows from the model’s design: Clairic does not generate conventional text output token by token. It returns values in the requested structure.
This pricing could be attractive for applications that process large volumes of documents or repeatedly request multiple fields. However, total operating costs also depend on infrastructure, application logic, validation and any other models used in the workflow.
The current preview is an early checkpoint. Cloudflare says it expects to improve the model’s arithmetic, estimation and forecasting capabilities as development continues.
What Developers Should Know About Accuracy and Reliability
Clairic’s structured output is intended to make results easier for applications to handle, but a correctly formatted result is not necessarily a correct answer.
For example, an estimated repair cost may fall within the developer’s specified range while still being wrong. A calculated deadline may use a misidentified date. A forecast may return a plausible number without accounting for a holiday or sudden change in demand.
Cloudflare provides confidence scores and a probability that a requested value is absent from the input. Estimates and forecasts also include uncertainty intervals. These signals can help developers determine when to accept a result, request another check or escalate a task for manual review.
For consequential applications, developers should combine model outputs with validation rules and independent evidence. Financial calculations can be handled by conventional code, while important invoice fields can be checked against source documents or business records.
This is especially important when a workflow could trigger a payment, reject a customer claim, change an account or make another consequential decision.
Clairic’s value is therefore not that it eliminates errors. It offers a different way to structure extraction work so developers can evaluate and control outputs more systematically.
The Bigger Picture
Cloudflare’s Clairic launch reflects a broader trend toward specialised AI models that handle defined tasks rather than relying on general-purpose systems for everything. Extraction models could reduce the cost and complexity of applications that need specific values from messy documents, while decision and generative models handle subsequent choices and communication.
The model also highlights an important distinction between structured output and verified truth. Typed fields, confidence scores and uncertainty intervals can make results easier to use, but they do not guarantee accuracy. Clairic’s success will depend on its performance on real workloads, the reliability of its estimates and forecasts, and whether developers can use it to simplify existing processing pipelines.
Looking Ahead
Cloudflare has released Clairic as a preview through Workers AI and plans to make the model’s weights openly available in a future release. Developers can begin evaluating the hosted model for extraction and lightweight forecasting tasks, but should account for the current limitations around complex calculations and longer-range predictions. Wider adoption will depend on how the model performs beyond the company’s initial examples.
The larger question is whether specialised extraction models can take repetitive workloads away from more expensive general-purpose language models. If they deliver dependable results at low cost, they could become useful components in document-processing and agentic workflows. For now, Clairic represents an early step in that direction, with validation and human oversight still necessary for important decisions.
Frequently Asked Questions
1. What is Cloudflare Clairic?
Clairic is a specialised AI model that extracts, computes, estimates and forecasts values from unstructured inputs. It returns structured values rather than conventional generated text.
2. How much does Clairic cost?
Cloudflare lists the preview price at $0.10 per million input tokens, with no output-token charges.
3. Is Clairic open-source or open-weight?
Cloudflare says it intends to release the model’s weights openly in the future. The October 19 launch provides hosted preview access, so the model should not yet be described as openly released.
4. What can Clairic extract from documents?
It can return values such as names, order IDs, dates, times, durations, numbers and lists. It can also compute certain values, estimate unknown quantities and forecast future values from historical data.
5. Can Clairic replace traditional forecasting systems?
Not in every case. Cloudflare describes its forecasting as basic and cautions that longer-horizon forecasts may be unreliable. Dedicated time-series models remain more appropriate for specialised or critical forecasting tasks.
6. How can developers access Clairic?
Developers can use the model through Cloudflare Workers AI with the identifier @cf/cloudflare/clairic, either through a Workers binding or the REST API.
Get the day’s top stories in your inbox
One concise email. No spam, unsubscribe anytime.
