AiGanakCheapest AI for document data extraction, by model and language
Extract fields from documents: what a typical job costs on every current OpenAI, Anthropic and Google model, in English, Hindi, Tamil and German. Pulling named fields into a fixed format is a narrow task. Small models usually handle it once the format is checked.
Extract fields from documents: a typical job is 2,250 tokens per document. At list prices checked 9 November 2026, the lowest-cost small-class model is GPT-6 Luna in English ($0.325 per 1,000 documents) and GPT-6 Luna in Hindi ($0.389 per 1,000 documents). Costs shown for English, Hindi, Tamil and German on every current OpenAI, Anthropic and Google model.
The pick, by language
Small class is the smallest model class that normally handles this job. The pick is the lowest-cost current model of that class or above.
| Language | Lowest-cost model | Per 1,000 documents | Per month at 10,000 documents |
|---|---|---|---|
| English | GPT-6 Luna OpenAI | $0.325 | $3 |
| Hindi | GPT-6 Luna OpenAI | $0.389 | $4 |
| Tamil | GPT-6 Luna OpenAI | $0.462 | $5 |
| German | GPT-6 Luna OpenAI | $0.362 | $4 |
Cost per 1,000 documents
Typical job: 500 tokens of instructions, 1,500 read and 250 written per document, counted in English tokens on the reference model family and converted for each model family and language with the measured token tax (4 November 2026). Standard list prices, no caching, no batch discount.
| Model | English | Hindi | Tamil | German |
|---|---|---|---|---|
| GPT-6 Luna OpenAI · Small | $0.325 | $0.389 | $0.462 | $0.362 |
| Claude Haiku 5.5 Anthropic · Small | $0.477 | $0.759 | $0.964 | $0.695 |
| Gemini 3.5 Flash-Lite Google · Small | $1.30 | $1.38 | $1.47 | $1.43 |
| Gemini 3.8 Flash Google · Mid | $2.58 | $2.79 | $3.01 | $2.92 |
| GPT-6.1 Terra OpenAI · Mid | $3.25 | $3.89 | $4.62 | $3.62 |
| GPT-6.1 Sol OpenAI · Mid | $6.50 | $7.78 | $9.24 | $7.24 |
| Gemini 4 Argon Google · Flagship | $6.88 | $7.76 | $8.94 | $7.74 |
| Gemini 3.1 Pro (preview) Google · Flagship · older or preview | $7.41 | $7.98 | $8.55 | $8.32 |
| Claude Sonnet 5.5 Anthropic · Mid | $9.54 | $15.18 | $19.29 | $13.91 |
| Claude Opus 5.5 Anthropic · Flagship | $19.08 | $30.36 | $38.58 | $27.82 |
| GPT-6 Astra OpenAI · Flagship | $32.50 | $38.91 | $46.20 | $36.20 |
| Claude Fable 5.1 Anthropic · Flagship | $47.69 | $75.90 | $96.44 | $69.54 |
Announced price changes are not included here: Gemini 4 Argon from 1 December 2026; Gemini 3.8 Flash from 1 January 2027. AiGanak Check applies them. The size of a job is an estimate; your own prompts and replies decide your bill. Quality on your task decides whether a cheaper model is usable.
Extract fields: questions
Can't find it? Press ⌘K and ask.
Which AI model is cheapest for document data extraction?
For a typical job of 500 instruction tokens, 1,500 tokens read and 250 tokens written per document, the lowest-cost small-class model at list prices checked 9 November 2026 is GPT-6 Luna in English at $0.325 per 1,000 documents. Smaller models cost less but may not reach the quality this job needs.
Does document data extraction cost more in Hindi or Tamil?
Usually yes, because the same text takes more tokens. The lowest-cost small-class model is GPT-6 Luna in Hindi at $0.389 per 1,000 documents and GPT-6 Luna in Tamil at $0.462 per 1,000 documents, against $0.325 in English. Instructions stay in English in these figures.
How is the cost per document worked out?
Tokens read and written per document are multiplied by each model's list price per million tokens, using the measured token counts of that model family for each language (AiGanak token tax index, 4 November 2026). Caching, batch discounts and retries are not included; AiGanak Check lets you turn them on.
Price another job
Support replies · Classify messages · Summarise · Translate · Research reports · Coding help · Document Q&A · Something else
Your real documents, priced on every model
The tables use a typical job. Your bill depends on your prompts, reply lengths and mix of languages. Send one month of usage and a sample of requests, and we price your actual traffic on every model and show the changes that are safe to make.
Delivered in two working days. If the savings identified come to less than USD 149 over three months, we refund what you paid.
Get an email when a price change affects these costs
We check OpenAI, Anthropic and Google pricing every week and email you the old price, the new price and the date.