Cloudflare 发布文档抽取模型 Clairic，官方称比大语言模型更快更便宜
Cloudflare's Clairic pulls data from documents faster and cheaper than LLMs
- 1Cloudflare 发布文档抽取模型 Clairic，Workers AI 可预览，权重将开源
- 2模型不做逐 token 生成，一次推理返回全部字段与置信度，缺失值返回 null
- 3官方称演示链路 19.3 秒完成，GLM-5.3 单独做三步需 61.7 秒、费用近两倍
- 4官方承认该模型尚不能可靠处理较长的计算链，建议此类计算放在自有代码中
Cloudflare 以专用模型承接字段抽取，实质是把智能体链路里高耗时高成本的通用推理环节换成低成本专用件。
- 影响：文档密集流程的抽取环节时延与成本有直接压缩空间；智能体链路评估需纳入专用模型加语言模型的混合组合；长链计算不可靠意味着复杂逻辑仍由自有代码承担。
- 参考：可考虑选一类票据或工单抽取场景接入 Clairic 预览版计时计价对比；值得把智能体流程拆成抽取、判断、生成三层再分别选型；长链计算暂留自有代码，不并入该模型。
- 依据：链路 19.3 秒对比 GLM-5.3 的 61.7 秒且费用近两倍，均为 Cloudflare 官方口径，暂无独立验证；长链计算不可靠也出自官方自述。
Cloudflare has released an AI model it says pulls data from documents faster and more cheaply than large language models. Clairic is available in preview on Workers AI, and open weights are set to follow soon.
Users name the fields they want, like an order number or invoice total, and Clairic returns matching values in the right format along with a confidence score. Unlike language models, Clairic doesn't generate text token by token. It answers every field in a single pass. If a value is missing, the model returns "null." Beyond extraction, it can handle simple calculations, estimates, and forecasts.
Cloudflare wants Clairic to run in chains alongside LLMs and other "System One" models like Jev. With a complaint email, for example, Clairic pulls out the order number, amount, and damage, then estimates the repair cost. The decision model Clef applies company rules to choose between a refund and a repair, and GLM-5.3 writes the reply to the customer.
Cloudflare says this chain takes 19.3 seconds. GLM-5.3 on its own needs 61.7 seconds for all three steps and costs almost twice as much.
According to Cloudflare, Clairic can't yet reliably handle longer chains of calculations. For now, developers should run those in their own code.
AI News Without the Hype – Curated by Humans
Subscribe to THE DECODER for ad-free reading, a weekly AI newsletter, our exclusive "AI Radar" frontier report six times a year, full archive access, and access to our comment section.
来源：The Decoder：AI News · the-decoder.com