Cloudflare releases Clairic, a single-pass document extraction model, in preview on Workers AI
Cloudflare has released Clairic, a model that extracts named fields from unstructured text in a single pass instead of generating tokens one at a time. It is in preview on Workers AI, with open weights promised soon. Cloudflare claims a multi-step agent chain using Clairic runs in 19.3 seconds, versus 61.7 seconds for GLM-5.3 alone.
Cloudflare has released Clairic, a document extraction model that answers every requested field in a single pass rather than generating text token by token. The model is available in preview on Workers AI, and Cloudflare says open weights will follow soon. No release date for the weights has been given.
How Clairic works
Users specify the fields they want, such as an order number or invoice total. Clairic returns a matching value for each field in the correct format, along with a confidence score. If a value is not present in the source text, the model returns null.
Because it is not an autoregressive text generator, Clairic does not produce free-form output. Beyond extraction, Cloudflare says it can handle simple calculations, estimates, and forecasts over the extracted values. It works on unstructured text and is designed as one component in a larger agent pipeline.
Chained with other "System One" models
Cloudflare positions Clairic to run alongside LLMs and other task-specific "System One" models, including one called Jev. Its example is a customer complaint email:
- Clairic extracts the order number, amount, and damage, then estimates the repair cost.
- Clef, a decision model, applies company rules to choose between a refund and a repair.
- GLM-5.3 writes the reply to the customer.
According to Cloudflare, this chain completes in 19.3 seconds. It says GLM-5.3 handling all three steps alone takes 61.7 seconds, roughly 3.2x longer, and costs almost twice as much. These figures come from Cloudflare and have not been independently verified. The workload, token counts, and hardware behind the comparison were not detailed.
Known limitations
Cloudflare says Clairic cannot yet reliably handle longer chains of calculations. For now, it advises developers to run those steps in their own code rather than relying on the model.
Specifications
| Item | Detail |
|---|---|
| Availability | Preview on Workers AI |
| Open weights | Announced, no date |
| Parameter count | Not disclosed |
| Context window | Not disclosed |
| Pricing | Pricing not yet disclosed |
| Training cutoff | Not disclosed |
| Published benchmark scores | None; only Cloudflare's latency and cost comparison |
What this means
Clairic is a bet that much of what teams use LLMs for in document workflows, such as pulling fields, normalizing formats, and flagging missing values, does not need autoregressive generation. A single-pass design gives predictable latency and per-field confidence scores, which matter in production pipelines where a silent hallucination is costlier than a null.
The real test is accuracy, and Cloudflare has published no extraction benchmarks, parameter count, or context window. The 19.3 versus 61.7 second comparison measures a chain in which an LLM does all the work against one that offloads two steps, so it says little about Clairic's extraction quality in isolation. Teams should wait for the open weights and independent evaluations on their own document types before replacing an LLM-based extractor.
The stated limit on multi-step calculations also means Clairic suits narrow, well-defined extraction tasks, not general reasoning over documents. If the open weights ship with permissive terms, it could become a cheap default for the extraction stage of agent pipelines.
Related Articles
MiniMax M3.1 Flash scores 42 on AA Intelligence Index at $0.15/$0.60 per 1M tokens, open weights Oct 23
MiniMax has released M3.1 Flash, a reasoning model that scores 42 on the Artificial Analysis Intelligence Index, up from 29 for M3. It costs $0.15 per 1M input tokens and $0.60 per 1M output tokens, and open weights are planned for October 23.
Mistral releases Timestral-209M, a time series model claiming top zero-shot scores on three benchmarks
Mistral AI has released Timestral-209M, the largest of four time series foundation models (5M, 28M, 123M, 209M parameters). According to the model card, it leads GIFT-Eval, TIME and fev-bench in zero-shot forecasting, with a context of up to 32,768 time steps and 21-quantile probabilistic output.
Tencent releases Hy-Memory-30B-A3B, a 3B-active model that writes agent memory for Mem0 and MemOS
Tencent Hunyuan released Hy-Memory-30B-A3B, a 30B-parameter mixture-of-experts model with about 3B active parameters, built to run the write path in external memory frameworks. Tencent reports a 55.49 cross-harness average, ahead of Claude Opus 5 (54.62) and DeepSeek V4.1 Flash (53.64) on its own fixed 20% benchmark subsamples.
DeepSeek V4.1 Flash 1017 launches with 552B CED backbone, 1M context, $0.15/$0.60 per 1M tokens
DeepSeek has released V4.1 Flash 1017, an October 17, 2026 snapshot built on its new Causal Encoder-Decoder architecture. It uses a 552B-parameter sparse MoE backbone with a 1M-token context window, native image understanding, and pricing of $0.15 input / $0.60 output per 1M tokens on OpenRouter.
Comments
Loading...