11-19 更新 实时
首页/ 技术栈/ 正文

Clairic 与 Clairic-flash 开放权重

技术栈 来源:Cloudflare Blog 2026-11-19 06:00 0 阅读 机器翻译
Clairic 与 Clairic-flash 开放权重

Cloudflare 已经成为一家 neolab。我们正在为客户和开源社区构建自己的前沿 AI 模型。

今天,我们宣布发布两个开放权重模型:

这两个抽取模型均托管在 Workers AI 上,完整的开放权重可在 Hugging Face 上获取。

我们也很高兴重新介绍自己:我们是 Workers AI 超级智能团队,也就是 Clef、Clairic 以及更多在研项目背后的前沿 AI 研究团队。这个团队正在快速推进 AI 前沿,为 Cloudflare 的客户和开源社区打造实用的模型。

自从以预览形式发布 Clairic 以来,我们收到了大量用例,以及关于客户希望如何使用抽取模型的有用反馈。我们听到的用例从收据等图像中抽取数据,到从支持工单中读取数值,不一而足。我们知道我们希望抽取成为该模型明确的重点,因此它必须在读取非结构化数据方面表现出色。

我们决定与 Clairic 一起推倒重来,重新设计架构,让这个新版本 Clairic 有别于预览版。对于普通的文本和数字抽取字段,我们现在采用并行槽位自回归生成,而非单次前向传播。每个字段利用自己已经写出的 Token 来选择下一个 Token,然后在各自的结束标记处停止。所有字段在服务批次中同步推进,每个请求的延迟依然极快。这提升了我们在文本和图像抽取方面的准确率,而准确率上的取舍也弥补了那一点延迟代价。对于计算、估算和抽取用例,我们训练了专门的 head,可处理估算、预测、时长、计算日期和显式算术运算。

尽管模型架构有所演进,我们仍保留了类型化 API 接口中让回答真正有用的那些部分:每次请求依然会返回置信度分数、字符跨度和概率。模型依然极快、极便宜。从下面的评估结果可以看到,我们较新版本的 Clairic 在文本、图像、PDF 等各类内容的抽取上都有了大幅提升。

Clairic-preview

Clairic-flash

全新文本字段,96 个请求

内部评估(mchen)——混合输入字段,36 个请求

已追踪的客户断言,139 种请求形态

文本和图像中的合成收据字段

目前还没有公开的抽取基准测试,但我们针对市场上其他新兴的抽取模型跑了自己的内部评估,例如 Liquid AI 的 D1.2(与 Apple Vision 流水线配合做光学字符识别)以及 Scaledown 的 Extract。我们的 Clairic 模型在各类抽取用例中都表现出色,不过这些评估属于自报数据,只能作为方向性参考。我们想借此展示 Clairic 在通用抽取用例中的表现概貌,同时也乐见其他公司也在发布抽取模型——这印证了我们最初的判断:由于成本、延迟和准确率问题,抽取一直是 LLM 用户的痛点。

我们对 Clairic 和 Clairic-flash 进行了延迟测试,测量的是自托管服务器返回的完整 HTTP 响应,包含请求中的所有字段。即便采用了重新设计的架构,你仍可预期响应时间低于一秒——对于类似的工作负载,这比典型的 LLM 响应要快得多。

各模型的请求数

Clairic p50 / p95

Clairic-flash p50 / p95

新文本字段

358 / 736 ms

155 / 328 ms

新图像字段

453 / 976 ms

305 / 605 ms

商业租赁

691 / 803 ms

379 / 519 ms

贷款定价

395 / 407 ms

188 / 192 ms

流量预测

532 / 585 ms

243 / 261 ms

费用审计、开发

632 / 742 ms

393 / 512 ms

在我们的预览博客文章中,Clairic 读取了一张收据照片。现在,Clairic 系列模型也能读取 PDF 了,因此你可以在一次请求中询问供应商、明细项和总额,即使它们出现在不同页面上。你总共最多可以包含四张图像或 PDF 页面,明细项可以以行的形式返回,将每项描述、数量和价格保持在一起。得益于 Clairic 的新架构,图像读取准确率在我们的内部基准测试中有所提升,错误从 207 个降至 0 个,明确覆盖文本和图像。

我们最初在 @cf/cloudflare/clairic 推出 Clairic 的预览端点时,托管的是 9B 版本,因为它训练更快、更容易,预览价格为每百万输入 Token 0.10 美元。我们现在将 9B 模型迁移到 @cf/cloudflare/clairic-flash 端点,而 27B 模型将位于 @cf/cloudflare/clairic。

27B 的 Clairic 模型定价为每百万输入 Token 0.24 美元,而 9B 的 Clairic-flash 模型定价为每百万输入 Token 0.10 美元。尽管该模型在技术层面确实会生成输出 Token,但我们决定沿用 System One 的定价原则,仅按输入 Token 收费:输出 Token 的生成成本由我们自己承担。

为了实现无缝切换,我们为 Clairic 提供促销折扣——在 2026 年 12 月 1 日之前,Clairic 和 Clairic-flash 均为每百万输入 Token 0.10 美元。这意味着你可以用同样的价格试用质量更高的 Clairic 版本。如果你的应用已经接入我们此前在预览版中发布的 Clairic 端点,你将自动受益于新模型更强的抽取能力,我们正在原地逐步推送这一升级。

如果你想将更注重成本的工作负载迁移到更便宜的 9B Clairic-flash 版本,我们鼓励你这样做。12 月 1 日,我们将结束 Clairic 的促销折扣,价格将恢复为每百万输入 Token 0.24 美元,而非折扣价每百万输入 Token 0.10 美元。

两款 Clairic 模型现已在 Hugging Face 上以开放权重形式发布。自我们发布 Clef 以来,我们看到大量实验和微调模型涌现,我们很高兴能通过 Clairic 的开放权重发布,继续为团队和研究提供支持。

我们推出这些 System One 模型的使命,是帮助人们用更小、更快、更便宜且针对特定任务专门构建的模型,来补充其昂贵的 LLM 使用。我们希望以开放权重形式发布这些模型,能鼓励更多人在我们的思路上开展实验并进一步构建。

Clairic 和 Clairic-flash 现可在 Workers AI 上通过 @cf/cloudflare/clairic 和 @cf/cloudflare/clairic-flash 使用。开发者文档中提供了完整的 API schema,这些 schema 与我们预览端点相比保持不变。我们还将在 Hugging Face 上发布 Clairic 和 Clairic-flash 的开放权重版本。

这些模型背后的团队——Workers AI 超级智能团队——才刚刚起步,并正在快速扩张,以推动 Cloudflare 的前沿 AI 研究。尽早加入我们,帮助塑造超级智能团队未来的模样。

译文由英文原文机器翻译,术语按行业惯例统一,可在文末展开原文核对。

查看英文原文(An updated Clairic and Clairic-flash are now open-weight, from the new Workers AI Superintelligence team)

Cloudflare has become a neolab. We’re building our own frontier AI models for our customers and the open-source community.

Today, we’re announcing two open-weight model releases:

  • We’re introducing an updated version of Clairic. Since the preview, we’ve improved extraction from complex documents including images and PDFs (new!), with more reliable handling of tables, lists, and dates, and better grounding of answers in the source material.
  • We’re introducing Clairic-flash, a lighter, faster version of Clairic that delivers strong extraction and prediction performance in a compact, 9-billion-parameter model.

Both extraction models are hosted on Workers AI and full open weights are available on Hugging Face.

We’re also excited to re-introduce ourselves as the Workers AI superintelligence team, the frontier AI research team behind Clef, Clairic, and lots more in the works. This team is working quickly to push the AI frontier and build useful models for Cloudflare’s customers and the open source community.

A new version of Clairic: redesigned after the preview release

Since announcing Clairic in preview, we have received a ton of use cases and helpful feedback on how customers wanted to use an extraction model. We heard about use cases ranging from extracting data from images like receipts to reading values from support tickets. We knew that we wanted extraction to be the clear focus for the model, so it had to excel at reading from unstructured data.

We decided to go back to the drawing board with Clairic and redesigned our architecture, differentiating this new version of Clairic from our preview release. For ordinary text and number extraction fields, we now use parallel slot autoregressive generation instead of a single forward pass. Each field uses the tokens it has already written to choose its next token, then stops at its own end marker. All the fields advance together in the serving batch, still giving lightning-fast latency per request. This boosted our accuracy in extraction for both text and images, and the trade-off in accuracy offsets the small latency cost. For computation, estimation, and extraction use cases, we trained special heads that can handle estimates, forecasts, durations, calculated dates, and explicit arithmetic.

Although the model architecture evolved, we kept the parts of the typed API interface that make the answers useful: confidence scores, character spans, and probabilities are still intact with every request. The model is still incredibly fast and cheap. From the evals below, you can see how our newer versions of Clairic have greatly improved on extraction across text, images, PDFs, and more.

Eval

Clairic-preview

Clairic

Clairic-flash

D1.2 + OCR

ScaleDown

Fresh text fields, 96 requests

295/336

335/336

328/336

304/336

259/336

Internal eval (mchen) - mixed-input fields, 36 requests

207/226

221/226

213/226

116/226

156/226

Tracked customer assertions, 139 request shapes

1,033/1,263

1,263/1,263

1,263/1,263

973/1,263

653/1,263

Synthetic receipt fields in text and images

369/576

576/576

576/576

397/576

280/576

There are no public extraction benchmarks yet, but we ran our internal evals for other emerging extraction models on the market, such as Liquid AI’s D1.2 (pipelined with Apple Vision for optical character recognition) and Scaledown’s Extract. Our Clairic models excel across extraction use cases, but take these evals as directional guidance since they are self-reported. We wanted to show an illustrative picture of how Clairic is performing across generalized extraction use cases and celebrate the fact that other companies are releasing extraction models as well — confirming our original hypothesis that extraction was a pain point for LLM users because of the cost, latency, and accuracy.

But is it fast?

We ran latency tests for Clairic and Clairic-flash, measuring complete HTTP response from a self-hosted server, including all fields in the request. Even with the redesigned architecture, you can expect responses in under one second — much faster than typical LLM responses for similar workloads.

Eval set

Requests per model

Clairic p50 / p95

Clairic-flash p50 / p95

Fresh text fields

96

358 / 736 ms

155 / 328 ms

Fresh image fields

48

453 / 976 ms

305 / 605 ms

Commercial leases

180

691 / 803 ms

379 / 519 ms

Loan pricing

600

395 / 407 ms

188 / 192 ms

Traffic forecasting

114

532 / 585 ms

243 / 261 ms

Expense audit, development

60

632 / 742 ms

393 / 512 ms

Better extraction for images and PDFs

In our preview blog post, Clairic read a photo of a receipt. Now the Clairic family of models can read PDFs too, so you can ask for the supplier, line items, and total in one request, even when they appear on different pages. You can include up to four images or PDF pages in total, and line items can come back as rows that keep each description, quantity, and price together. Due to Clairic’s new architecture, image reading accuracy has improved from 207 errors to 0 on our internal benchmarks, explicitly covering text and images.

Promotional pricing on Clairic for a limited time

When we first launched our preview endpoint of Clairic at @cf/cloudflare/clairic, we hosted the 9B variant as it was faster and easier to train, at the preview price of $0.10 per M input tokens. We’re now moving the 9B model to the @cf/cloudflare/clairic-flash endpoint and the 27B model will live at @cf/cloudflare/clairic.

The 27B Clairic model will be priced at $0.24 per M input tokens, while the 9B Clairic-flash model will be priced at $0.10 per M input tokens. Although the model does technically generate output tokens, we decided to keep the System One pricing principles and are only charging on input tokens: we eat the cost of the output token generation.

In order to make this a seamless swap, we’re offering a promotional discount on Clairic — until December 1, 2026, both Clairic and Clairic-flash will cost $0.10 per M input tokens. This means you get to try out the higher quality variant of Clairic at the same price. If your applications are hooked up to the Clairic endpoint that we released in preview, you will automatically benefit from the better extraction capabilities of our new model, which we’re rolling out in place.

If you’d like to move more cost-sensitive workloads to the cheaper 9B Clairic-flash variant, we encourage you to do so. On December 1, we’ll be ending the promotional discount for Clairic and pricing will resume at $0.24 per M input tokens instead of the discounted $0.10 per M input tokens.

Clairic and Clairic-flash (new) are open-weight

Both Clairic models are now open-weight on Hugging Face as well. We have seen an outpouring of experiments and tuned models from our release of Clef, and we’re happy to continue supporting teams and research with our open-weight release of Clairic.

Our mission with these System One models is to help people augment their expensive LLM usage with smaller, faster, cheaper models that are purpose-built for a specific task. We hope that releasing these models open-weight encourages others to experiment and build on our ideas as well.

Try them out

Clairic and Clairic-flash are available on Workers AI as @cf/cloudflare/clairic and @cf/cloudflare/clairic-flash. The developer documentation has the full API schemas, which remain unchanged from our preview endpoint. We’re also publishing the open-weight releases for Clairic and Clairic-flash on Hugging Face. 

The team behind these models — the Workers AI superintelligence team — is just getting started, and growing fast to push frontier AI research at Cloudflare. Join us early and help shape what the superintelligence team becomes.

正文保留英文原文,标题与摘要由机器翻译。

本文由采集器自动同步

原文发布于 Cloudflare Blog(An updated Clairic and Clairic-flash are now open-weight, from the new Workers AI Superintelligence team),标题、摘要与正文由机器翻译,术语按行业惯例统一,未经人工逐句校订。

阅读原文 ↗