← Nikhil Mudholkar

Independent test · 21 October 2026 · Clairic preview

Clairic, two days in

On 19 October Cloudflare released Clairic, a new kind of model that reads documents and fills in fields. We tested the preview on 503 real order emails and 711 receipt photos, next to five LLMs. It is the fastest and most consistent model we tested. It is not yet the most reliable one.

503real order emails
711receipt photos
6models
~30,000API calls

This is a two-day-old preview. Cloudflare says Clairic will improve with each checkpoint, and we expect it to. Everything below describes the hosted model on 20-21 October 2026. We will run the same tests on the next version.

The model

What Clairic does

An LLM writes its answer one word at a time. Clairic does not write. You give it a document and a list of named fields, and it returns one typed value per field in a single pass: a piece of text, a number, a date. For each value it also returns a confidence and a probability that the value is not in the document at all. Cloudflare calls this a "System One extraction model".

The launch post says it is faster, cheaper and more accurate than LLMs for this kind of work, and that it returns empty instead of guessing when a value is missing. We tested those claims on two jobs that matter for real automation: reading orders from customer emails, and reading totals from receipt photos.

ClairicLLMs

Finding 01

Where Clairic is strong

Speed and consistency are real. Clairic was the fastest model in every test, and it gave the same answer to the same input almost every time.

Clairic answers fastest

Median time per email, 100 emails, one model at a time. Lower is better.

0 s1.0 s2.0 s3.0 s4.0 sClairic: 0.36 s medianClairic0.4 sGemini 3.1 Flash-Lite: 1.03 s medianGemini 3.1 Flash-Lite1.0 sClaude Haiku 5.5: 1.38 s medianClaude Haiku 5.51.4 sClaude Sonnet 5.5: 1.82 s medianClaude Sonnet 5.51.8 sGemini 3.8 Flash: 1.90 s medianGemini 3.8 Flash1.9 sGLM-5.3: 3.84 s medianGLM-5.33.8 s

Sequential calls from one GCP VM · 2026-10-20

Clairic gives the same answer every time

Share of answers that changed when the same 100 emails were sent three times. Lower is better.

0%5.0%10.0%15.0%20.0%Clairic: 0.2% of answers changedClairic0.2%Gemini 3.8 Flash: 0.9% of answers changedGemini 3.8 Flash0.9%Claude Sonnet 5.5: 1.2% of answers changedClaude Sonnet 5.51.2%Claude Haiku 5.5: 1.6% of answers changedClaude Haiku 5.51.6%GLM-5.3: 15.6% of answers changedGLM-5.315.6%Gemini 3.1 Flash-Lite: 18.1% of answers changedGemini 3.1 Flash-Lite18.1%

426 answers per model · each vendor's default sampling

It also reads values that are clearly printed. It read the printed total on 95% of receipts and the quantity on 97.6% of email lines. And when nothing on a receipt could tell the total, it said so on 99.6% of them, as did every LLM.

Finding 02 · Customer emails

It fills in values that are not there

We took 503 real request-for-quote emails from one of our customers, an industrial instruments distributor. For every ordered product, each model had to return the part number and the quantity, or nothing if the email did not state them. Two models labelled the answers independently, and we checked every disagreement by hand.

The test that matters for automation is the empty case. A wrong part number that looks right is worse than no answer, because it goes straight into a quote.

Asked for a part number the customer never wrote

Share of answers that returned a part number anyway. Lower is better.

0%25%50%75%100%Claude Sonnet 5.5: 3.1% of 515Claude Sonnet 5.53%Claude Haiku 5.5: 4.9% of 515Claude Haiku 5.55%Gemini 3.8 Flash: 6.8% of 515Gemini 3.8 Flash7%Gemini 3.1 Flash-Lite: 25.0% of 515Gemini 3.1 Flash-Lite25%GLM-5.3: 37.2% of 384GLM-5.337%Clairic: 67.6% of 515Clairic68%

515 product lines with no part number in 503 real order emails (GLM-5.3: 384)

We deleted the quantity. Who still gave one?

Share of answers with a quantity after it was removed from the email. Lower is better.

0%10%20%30%40%Gemini 3.8 Flash: 0.2% of 435Gemini 3.8 Flash0%Claude Sonnet 5.5: 0.5% of 435Claude Sonnet 5.50%GLM-5.3: 1.8% of 328GLM-5.32%Claude Haiku 5.5: 1.8% of 435Claude Haiku 5.52%Gemini 3.1 Flash-Lite: 2.1% of 435Gemini 3.1 Flash-Lite2%Clairic: 38.2% of 435Clairic38%

435 deletions (GLM-5.3: 328) · when other lines still showed quantities, Clairic: 54%

What Clairic returns instead is usually real text from nearby: the product description, a size such as "20-50mm", another line's code, or "2 Nos". Its probability that the value is missing stayed low on these answers (median 0.07 for part numbers), so the empty signal alone does not catch them.

Reading a part number that is there

Share of part numbers read exactly as the customer wrote them. Higher is better.

0%25%50%75%100%Gemini 3.8 Flash: 96.9% of 589Gemini 3.8 Flash97%Claude Sonnet 5.5: 95.8% of 589Claude Sonnet 5.596%Claude Haiku 5.5: 94.9% of 589Claude Haiku 5.595%Gemini 3.1 Flash-Lite: 94.7% of 589Gemini 3.1 Flash-Lite95%GLM-5.3: 94.6% of 463GLM-5.395%Clairic: 82.3% of 589Clairic82%

589 part numbers (GLM-5.3: 463) · quantities: Clairic 97.6%, all LLMs 98.9-99.9%

How much can you fill in without a human?

Largest share of answers you can accept, sorted by the model's own confidence, and still be right 99% of the time.

0%25%50%75%100%Claude Sonnet 5.5: 97.0%Claude Sonnet 5.597%Claude Haiku 5.5: 90.0%Claude Haiku 5.590%Gemini 3.8 Flash: 84.1%Gemini 3.8 Flash84%Clairic: 23.7%Clairic24%Gemini 3.1 Flash-Lite: 1.4%Gemini 3.1 Flash-Lite1%GLM-5.3: 1.0%GLM-5.31%

Part numbers and quantities on the unedited emails · higher is better

Finding 03 · Receipt photos

It reads a total, but does not work one out

We took 711 public receipt photos where the items, tax and total add up exactly. Then we put a yellow sticker over more and more of the amounts and asked every model for the total.

One receipt, four stickers

Every model got the same photo and the same question: what is the total? The real total is 283,338 rupiah.

Receipt, nothing covered

A · Nothing covered

Read the total

  • Clairic283,338
  • Haiku 5.5283,338
  • Gemini 3.8 Flash283,338
  • Sonnet 5.5283,338
  • Opus 5.5283,338
Receipt, total covered

B · Total covered

Subtotal + tax still visible

  • Clairic25,758
  • Haiku 5.5283,338
  • Gemini 3.8 Flash283,338
  • Sonnet 5.5283,338
  • Opus 5.5283,338
Receipt, total + subtotal covered

C · Total + subtotal covered

Add up the items + tax

  • Clairic257,558
  • Haiku 5.5283,338
  • Gemini 3.8 Flash283,338
  • Sonnet 5.5283,338
  • Opus 5.5283,338
Receipt, all amounts covered

D · All amounts covered

Right answer: can't tell

  • Clairiccan't tell
  • Haiku 5.5can't tell
  • Gemini 3.8 Flashcan't tell
  • Sonnet 5.5can't tell
  • Opus 5.5can't tell

CORD-v2 receipt validation_34 (public, CC-BY 4.0) · green = right, red = wrong

Right answers on 711 receipts

Each panel is one sticker version. Higher is better.

A · Read the printed total

0%25%50%75%100%Clairic: 94.9%Clairic95%Claude Haiku 5.5: 89.9%Haiku 5.590%Gemini 3.8 Flash: 99.7%Gemini 3.8 Flash100%Claude Sonnet 5.5: 99.6%Sonnet 5.5100%Claude Opus 5.5: 99.7%Opus 5.5100%

B · Total covered

0%25%50%75%100%Clairic: 53.6%Clairic54%Claude Haiku 5.5: 91.6%Haiku 5.592%Gemini 3.8 Flash: 97.3%Gemini 3.8 Flash97%Claude Sonnet 5.5: 98.2%Sonnet 5.598%Claude Opus 5.5: 99.4%Opus 5.599%

C · Total and subtotal covered

0%25%50%75%100%Clairic: 34.0%Clairic34%Claude Haiku 5.5: 90.4%Haiku 5.590%Gemini 3.8 Flash: 96.2%Gemini 3.8 Flash96%Claude Sonnet 5.5: 92.5%Sonnet 5.593%Claude Opus 5.5: 97.6%Opus 5.598%

D · Everything covered: says can't tell

0%25%50%75%100%Clairic: 99.6%Clairic100%Claude Haiku 5.5: 100.0%Haiku 5.5100%Gemini 3.8 Flash: 100.0%Gemini 3.8 Flash100%Claude Sonnet 5.5: 100.0%Sonnet 5.5100%Claude Opus 5.5: 100.0%Opus 5.5100%

711 receipts × 4 versions × 5 models · a sticker on an empty area changed no model by more than 1.2 points

With the total covered, the LLMs worked it out 90%-98% of the time. Clairic did 34%. Often it had the right number: its hidden best guess, which it returns beside the answer, was right on 62% of these receipts. But its own threshold turned that guess into "can't tell". Other times it reported the tax line or the item sum as the total.

Working out the total, by number of items

Version C: total and subtotal covered. Share right.

Model1 item (338)2 items (172)3-5 items (156)6+ items (45)
Clairic37%37%33%4%
Claude Haiku 5.584%95%98%93%
Gemini 3.8 Flash95%98%97%98%
Claude Sonnet 5.596%96%88%67%
Claude Opus 5.598%99%97%89%

Receipt counts in brackets

Finding 04

The launch claims

Cloudflare's launch claims, checked

Our measurement on this data, not a general verdict.

ClaimWhat we measuredResult
Faster than LLMsMedian 0.36 s per email against 1.03-3.84 sHolds
Same answer every run0.2% of answers changed over three runsHolds
Cheaper$0.31 per 1,000 emails against $0.25 for Claude Haiku 5.5; receipts $0.23 against $0.28Close to Haiku
Thirty fields cost about the same as three30 fields cost 12.5 times as much as 2: about one pass per field, each re-reading the emailNot on our data
Returns null when a value is not in the inputReturned a part number for 68% of lines that had noneNot on our data
One- and two-step sums are reliableTotal from subtotal + tax: 54% right, LLMs 92%-99%Not on our data
Use confidence to decide what to automateConfidence ranks answers well (0.89), but only 24% can be accepted at 99% precision; Haiku 90%Partly

Claims from the launch post, 2026-10-19 · preview checkpoint

Practical advice

How we would use it today

Method

How we tested

Nikhil Mudholkar · Bryo · independent test, not sponsored by any vendor.

© 2026 Nikhil Mudholkar