Cheapest LLM for batch classification and extraction
Offline jobs such as tagging, routing, moderation or structured extraction rarely need an answer in seconds. That makes batch discounts the first thing to check. Prices verified 2026-10-18.
Assumptions
- Classification: 1,000,000 documents, 600 input and 20 output tokens each (a label)
- Extraction: 100,000 documents, 3,000 input and 500 output tokens each (a JSON object)
- Batch API pricing where the provider offers it; standard pricing otherwise (tagged "no batch API")
- No prompt caching (instructions are short relative to each document)
Change any of these in the calculator: every row below has a link that opens the same scenario for that model.
The short answer
- Cheapest flagship model: Mistral Large 4 at $225 (temporary sale price; regular $1.36 / $4.18 per 1M)
- Cheapest mid-tier model: Gemini 3.8 Flash at $263 (price rises 2027-01-01)
- Cheapest small model: Claude Haiku 5.5 at $35.00
Output-heavy jobs rank differently
Classification is almost all input, so input price decides. Extraction writes 500 output tokens per document, so output price starts to matter. The five cheapest models for extraction:
- Ministral 3 8B: $26.25 per 100K documents
- Claude Haiku 5.5: $27.50 per 100K documents
- GPT-6 Luna: $27.50 per 100K documents
- Mistral Small 4: $37.50 per 100K documents
- GPT-5.6 Luna: $60.00 per 100K documents
DeepSeek off-peak instead of batch
DeepSeek has no batch discount but publishes off-peak rates. Running the same jobs in off-peak hours:
- DeepSeek V4.1 Flash: classification $102, extraction $75.00
- DeepSeek V4 Pro: classification $436, extraction $297
Trade-offs
Batch APIs return results asynchronously rather than in seconds; check each provider's batch documentation for the completion window. Small models are often accurate enough for clear-cut labels but miss more on ambiguous ones. A common pattern is to run the cheap model on everything, then send low-confidence items to a larger model. Measure accuracy on a labelled sample of a few hundred documents first: a 2% error rate on 1M documents is 20,000 wrong labels.
Full ranking (36 models)
| # | Model | Tier | Classify 1M docs | Extract 100K docs | |
|---|---|---|---|---|---|
| 1 | Claude Haiku 5.5 Anthropic | small | $35.00 | $27.50 | Calculator |
| 2 | GPT-6 Luna OpenAI | small | $35.00 | $27.50 | Calculator |
| 3 | Ministral 3 8B Mistral AI | small | $46.50 | $26.25 | Calculator |
| 4 | Mistral Small 4 Mistral AI | small | $51.00 | $37.50 | Calculator |
| 5 | GPT-5.6 Luna OpenAI | small | $72.00 | $60.00 | Calculator |
| 6 | Gemini 3.1 Flash-Lite Google | small | $90.00 | $75.00 | Calculator |
| 7 | Codestral Mistral AI | small | $99.00 | $67.50 | Calculator |
| 8 | Gemini 3.5 Flash-Lite Google | small | $115 | $108 | Calculator |
| 9 | Gemini 2.5 Flash Google | small | $115 | $108 | Calculator |
| 10 | DeepSeek V4.1 Flash DeepSeek no batch API | small | $204 | $150 | Calculator |
| 11 | Mistral Large 4 Mistral AI sale price | flagship | $225 | $154 | Calculator |
| 12 | Gemini 3.8 Flash Google price changes 2027-01-01 | mid | $263 | $206 | Calculator |
| 13 | Gemini 3.6 Flash Google price changes 2027-01-01 | mid | $263 | $206 | Calculator |
| 14 | GPT-5.4 mini OpenAI | small | $270 | $225 | Calculator |
| 15 | Claude Haiku 4.5 Anthropic | small | $350 | $275 | Calculator |
| 16 | Mistral Medium 3.5 Mistral AI | mid | $525 | $413 | Calculator |
| 17 | Grok 4.20 (Reasoning) xAI | reasoning | $640 | $400 | Calculator |
| 18 | Grok 4.3 xAI | mid | $640 | $400 | Calculator |
| 19 | Grok Build 0.1 xAI no batch API | small | $640 | $400 | Calculator |
| 20 | Gemini 4 Argon Google price changes 2026-12-01 | flagship | $700 | $550 | Calculator |
| 21 | Claude Sonnet 5.5 Anthropic | mid | $700 | $550 | Calculator |
| 22 | Claude Sonnet 5 Anthropic | mid | $700 | $550 | Calculator |
| 23 | GPT-6 Sol OpenAI | mid | $700 | $550 | Calculator |
| 24 | GPT-6.1 Sol OpenAI | mid | $700 | $550 | Calculator |
| 25 | Gemini 3.1 Pro (Preview) Google | flagship | $720 | $600 | Calculator |
| 26 | GPT-5.6 Terra OpenAI | mid | $720 | $600 | Calculator |
| 27 | DeepSeek V4 Pro DeepSeek no batch API | flagship | $871 | $594 | Calculator |
| 28 | Claude Sonnet 4.6 Anthropic | mid | $1,050 | $825 | Calculator |
| 29 | Grok 4.7 xAI no batch API | flagship | $1,320 | $900 | Calculator |
| 30 | Claude Opus 5.5 Anthropic | flagship | $1,400 | $1,100 | Calculator |
| 31 | GPT-5.6 Sol OpenAI | flagship | $1,400 | $1,100 | Calculator |
| 32 | Claude Opus 5 Anthropic | flagship | $1,750 | $1,375 | Calculator |
| 33 | Claude Opus 4.8 Anthropic | flagship | $1,750 | $1,375 | Calculator |
| 34 | GPT-5.5 OpenAI | flagship | $1,800 | $1,500 | Calculator |
| 35 | Claude Fable 5.1 Anthropic | flagship | $3,500 | $2,750 | Calculator |
| 36 | GPT-6 Astra OpenAI | flagship | $3,500 | $2,750 | Calculator |
USD, standard list prices. Tags mark announced price changes and temporary sale prices.
Price is not quality
This page ranks by cost only. Models differ in accuracy, instruction following, tool use, latency, rate limits and data policies, and those differences often matter more than the price gap. Shortlist two or three models from the table, run them on a sample of your real inputs, and compare cost per successful result.
Related comparisons
More guides
- Cheapest LLM for coding agents
- Cheapest LLM for RAG chatbots
- Long-context document analysis (400K tokens)
- Cheapest LLM for high-volume customer support
- Prompt caching explained, with savings
FAQ
What is the cheapest way to classify a million documents with an LLM?
With the assumptions here, Claude Haiku 5.5 costs $35.00 for 1,000,000 documents. Shorter prompts and single-token labels lower it further.
How big are batch discounts?
Most providers in this table list 50% off for batch requests. Models without a published batch discount are tagged in the table.