WhichAI.today

Best AI for extracting data from invoices right now

Updated Oct 24, 2026. Checked daily.

Ask for the fields you need, one invoice at a time, and check the amounts against the file.

The speed board measures output speed, not extraction accuracy.

Try this prompt

Extract the supplier, date, and total from this invoice. If a field is missing, say missing.

Do not upload customer invoices to a consumer chat that trains on chats by default.

#1 Champion

Gemini 3.5 Flash-Lite

Google

Google AI Studio free tier, or the Gemini app, choose Gemini 3.5 Flash-Lite.

Free (limited)

Leads Artificial Analysis output speed

The fastest output of any tracked model, with a free API tier for prototyping.

Artificial Analysis output speed 412 tokens/sec (Oct 2026).

Open Google
Details
Why it leads
Artificial Analysis output speed 412 tokens/sec (Oct 2026). Source.
Cost
Free API tier with rate limits. Paid API is $0.30 per million input tokens and $2.50 per million output.
Free tier
Yes. Free API use can be used to improve Google products. The paid API tier is not.
Privacy default
Yes for adults. Gemini Apps Activity is on by default, and those chats can be used to improve Google AI. Gemini Apps Activity, turn off Keep Activity. Safety review can still apply. Source
When not to use it
Skip it for hard coding, legal reading, or anything where a miss is expensive.
Starter prompt
Classify each of these support messages as billing, bug, or other, one label per line.

Top Alternative

Claude Haiku 5.5

Anthropic

Claude app, free plan includes Haiku 5.5.

Free (limited)

Very fast while staying far smarter than most budget models, at a $0.10 per million token input price.

Artificial Analysis output speed 237 tokens/sec.

Open Anthropic
Details
Why it leads
Artificial Analysis output speed 237 tokens/sec. Source.
Cost
Included on the free plan, with limits. API is $0.10 per million input tokens for prompts under 100K tokens.
Free tier
Yes, on the free Claude plan, with usage limits.
Privacy default
Anthropic says Free, Pro, and Max chats are used for training only if you allow it, and you choose at signup. The toggle has been reported as on by default, so check it. Settings, then Privacy. Turn off Help improve Claude. Source
When not to use it
Skip it when the task is hard enough that Opus or Fable would change the answer.
Starter prompt
Reply to this customer in two sentences. Be specific and do not over-promise.

Best Open Source

DeepSeek V4.1 Flash

DeepSeek

DeepSeek chat is free. Choose V4.1 Flash, or download weights on Hugging Face.

Free (limited)

Fast, extremely cheap, MIT-licensed, and the smartest model in the budget tier.

Artificial Analysis output speed not cited. AA Intelligence Index 45.

Open DeepSeek
Details
Why it leads
Artificial Analysis output speed not cited. AA Intelligence Index 45. Source.
Cost
Chat is free. The API costs a fraction of a dollar per million tokens, with 50% discounts off-peak.
Free tier
Yes. The chat site and app are free, with throttling when the service is busy.
Privacy default
Possibly. DeepSeek says some optimization data can come from user input, after de-identification. Use the opt-out described in the DeepSeek privacy policy. Source
When not to use it
Skip the hosted chat for private data. The weights can be run on your own machines.
Starter prompt
Tag each row as yes or no, and keep the output to a single column.

⚡ More on Speed & High Volume