Cloudflare launches Clairic, a System One extraction model faster and cheaper than LLMs
Cloudflare researchers including Michelle Chen have released Clairic, a "System One"-style extraction model purpose-built to extract, compute, estimate, and predict numeric values from unstructured input. It is now live on Workers AI (model IDs follow the @cf/cloudflare… format) as a research preview, with an open-weights version to follow once the team finishes architecture and capability adjustments.
Confirmed
- Positioning and capabilities: Clairic is a "system one model" — it reads input in a single forward pass with no token generation, and outputs strongly typed values, confidence scores, probabilities, and character offsets (making it easy to check against the source text). It can also read images. Unlike decision models, it doesn't require a pre-enumerated list of possible answers.
- Pricing and cost: no per-output-token charges; its architecture keeps cost and latency far below those of LLMs.
- Benchmarks: the authors demonstrated a Clairic + Clef + LLM cascaded pipeline running 3.5x faster and 77% cheaper than an LLM alone.
- Release status: hosted on Workers AI; Neolab says the open-weights version will ship once capabilities and architecture meet expectations.
Why it matters
- Numeric extraction with LLMs is a well-known pain point — models hallucinate numbers that don't exist. Clairic offers an alternative path combining a small, specialized model with cascaded pipelines, delivering speed, cost, and reliability advantages for large-scale structured extraction.
2026-10-20 ~ 2026-10-20 · 8 related posts
Primary sources
- Cloudflare unveils Clairic, a System One extraction model claiming higher accuracy than LLMs — michellechen ·
- Clairic + Clef + LLM cascaded pipeline is 3.5x faster and 77% cheaper than an LLM alone — michellechen ·
- Clairic is live on Workers AI; open-weights release planned once the architecture matures — michellechen ·
- [source] Cloudflare unveils Clairic, a System One extraction model claiming higher accuracy than LLMs — michellechen · 2026-10-20
- Neolab unveils Clairic, a 'system one' extraction model that reads inputs in a single forward pass — michellechen · 2026-10-20
- Clairic's capability sheet: typed outputs, confidence scores, no output-token charges — michellechen · 2026-10-20
- [source] Clairic + Clef + LLM cascaded pipeline is 3.5x faster and 77% cheaper than an LLM alone — michellechen · 2026-10-20
- [source] Clairic is live on Workers AI; open-weights release planned once the architecture matures — michellechen · 2026-10-20
- Clairic debuts: a regression-based System 1 model, open weights coming soon — michellechen · 2026-10-20
2 near-duplicate retellings: ritakozlov · irvinebroque