Dataset last verified . How we verify
These are illustrative workloads, not measured usage or model recommendations. Rankings compare Standard list prices only, before taxes, negotiated discounts and extra service charges. Published context-dependent rates are selected for the disclosed input length. Equal costs are ordered by model ID.
Cheaper per token is not cheaper per successful task. Capabilities and task outcomes require your own evaluation.
At a glance
| Use case | Cheapest listed product | Monthly cost | Runner-up |
|---|---|---|---|
| High-volume classification | Ministral 3 3B | $5.20 | Claude Haiku 5.5 · $6.00 |
| Customer support chat | Ministral 3 3B | $13.25 | Claude Haiku 5.5 · $17.57 |
| RAG answers | Ministral 3 3B | $8.50 | Claude Haiku 5.5 · $10.50 |
| Coding agent task | Ministral 3 3B | $1.52 | Ministral 3 8B · $2.28 |
| Long-document summarization | Ministral 3 3B | $12.90 | GPT-6 Luna · $13.30 |
| OCR pages | Google Enterprise Document OCR | $13.50 | Amazon Textract DetectDocumentText · $15.00 |
High-volume classification
Example workload: 100,000 requests/month; 500 input and 20 output tokens/request; no cache reads assumed.
Selection: Active, verified text-output models with source-linked model pricing pages.
| Rank | Model / product | Provider | Monthly cost | Price inputs used | Official source | Last verified How we verify |
|---|---|---|---|---|---|---|
| 1 | Ministral 3 3B | Mistral AI | $5.20 | Input $0.10 / million tokens; output $0.10 / million tokens; Default rates. | Published pricing | |
| 2 | Claude Haiku 5.5 | Anthropic | $6.00 | Input $0.10 / million tokens; output $0.50 / million tokens; Short context rates. | Published pricing | |
| 3 | GPT-6 Luna | OpenAI | $6.00 | Input $0.10 / million tokens; output $0.50 / million tokens; Short context rates. | Published pricing | |
| 4 | Ministral 3 8B | Mistral AI | $7.80 | Input $0.15 / million tokens; output $0.15 / million tokens; Default rates. | Published pricing | |
| 5 | Mistral Small 4 | Mistral AI | $8.70 | Input $0.15 / million tokens; output $0.60 / million tokens; Default rates. | Published pricing |
89 products considered; 50 eligible across 8 providers; 39 excluded.
- No eligible active text-model pricing page: 14.
- Unverified pricing: 5.
- Missing token prices: 11.
- Missing source or verification date: 4.
- Non-text output: 5.
Customer support chat
Example workload: 10,000 conversations/month; 6 replies/conversation; 600 system, 90 user and 1,000 retrieved-context tokens/reply; 180 output tokens/reply; 50% history carry. Total per conversation: 12,165 input and 1,080 output tokens; no cache reads assumed.
Selection: Active, verified text-output models; the shared customer-support conversation and history assumptions apply.
| Rank | Model / product | Provider | Monthly cost | Price inputs used | Official source | Last verified How we verify |
|---|---|---|---|---|---|---|
| 1 | Ministral 3 3B | Mistral AI | $13.25 | Input $0.10 / million tokens; output $0.10 / million tokens; Default rates. | Published pricing | |
| 2 | Claude Haiku 5.5 | Anthropic | $17.57 | Input $0.10 / million tokens; output $0.50 / million tokens; Short context rates. | Published pricing | |
| 3 | GPT-6 Luna | OpenAI | $17.57 | Input $0.10 / million tokens; output $0.50 / million tokens; Short context rates. | Published pricing | |
| 4 | Ministral 3 8B | Mistral AI | $19.87 | Input $0.15 / million tokens; output $0.15 / million tokens; Default rates. | Published pricing | |
| 5 | Mistral Small 4 | Mistral AI | $24.73 | Input $0.15 / million tokens; output $0.60 / million tokens; Default rates. | Published pricing |
89 products considered; 50 eligible across 8 providers; 39 excluded.
- No eligible active text-model pricing page: 14.
- Unverified pricing: 5.
- Missing token prices: 11.
- Missing source or verification date: 4.
- Non-text output: 5.
RAG answers
Example workload: 10,000 requests/month; 8,000 input and 500 output tokens/request; no cache reads assumed.
Selection: Active, verified text-output models; retrieved context is uncached.
| Rank | Model / product | Provider | Monthly cost | Price inputs used | Official source | Last verified How we verify |
|---|---|---|---|---|---|---|
| 1 | Ministral 3 3B | Mistral AI | $8.50 | Input $0.10 / million tokens; output $0.10 / million tokens; Default rates. | Published pricing | |
| 2 | Claude Haiku 5.5 | Anthropic | $10.50 | Input $0.10 / million tokens; output $0.50 / million tokens; Short context rates. | Published pricing | |
| 3 | GPT-6 Luna | OpenAI | $10.50 | Input $0.10 / million tokens; output $0.50 / million tokens; Short context rates. | Published pricing | |
| 4 | Ministral 3 8B | Mistral AI | $12.75 | Input $0.15 / million tokens; output $0.15 / million tokens; Default rates. | Published pricing | |
| 5 | Mistral Small 4 | Mistral AI | $15.00 | Input $0.15 / million tokens; output $0.60 / million tokens; Default rates. | Published pricing |
89 products considered; 50 eligible across 8 providers; 39 excluded.
- No eligible active text-model pricing page: 14.
- Unverified pricing: 5.
- Missing token prices: 11.
- Missing source or verification date: 4.
- Non-text output: 5.
Coding agent task
Example workload: 1,000 tasks/month; 8 calls/task; 1,500 input and 400 output tokens/call; no cache reads assumed.
Selection: Active, verified text-output models; the shared coding-agent calls and token assumptions apply.
| Rank | Model / product | Provider | Monthly cost | Price inputs used | Official source | Last verified How we verify |
|---|---|---|---|---|---|---|
| 1 | Ministral 3 3B | Mistral AI | $1.52 | Input $0.10 / million tokens; output $0.10 / million tokens; Default rates. | Published pricing | |
| 2 | Ministral 3 8B | Mistral AI | $2.28 | Input $0.15 / million tokens; output $0.15 / million tokens; Default rates. | Published pricing | |
| 3 | Claude Haiku 5.5 | Anthropic | $2.80 | Input $0.10 / million tokens; output $0.50 / million tokens; Short context rates. | Published pricing | |
| 4 | GPT-6 Luna | OpenAI | $2.80 | Input $0.10 / million tokens; output $0.50 / million tokens; Short context rates. | Published pricing | |
| 5 | Ministral 3 14B | Mistral AI | $3.04 | Input $0.20 / million tokens; output $0.20 / million tokens; Default rates. | Published pricing |
89 products considered; 50 eligible across 8 providers; 39 excluded.
- No eligible active text-model pricing page: 14.
- Unverified pricing: 5.
- Missing token prices: 11.
- Missing source or verification date: 4.
- Non-text output: 5.
Long-document summarization
Example workload: 1,000 requests/month; 128,000 input and 1,000 output tokens/request; no cache reads assumed.
Selection: Active, verified text-output models whose published context window can hold the disclosed single-request input; unknown context capacity is excluded. Minimum published input capacity: 128,000 tokens.
| Rank | Model / product | Provider | Monthly cost | Price inputs used | Official source | Last verified How we verify |
|---|---|---|---|---|---|---|
| 1 | Ministral 3 3B | Mistral AI | $12.90 | Input $0.10 / million tokens; output $0.10 / million tokens; Default rates. | Published pricing | |
| 2 | GPT-6 Luna | OpenAI | $13.30 | Input $0.10 / million tokens; output $0.50 / million tokens; Short context rates. | Published pricing | |
| 3 | Ministral 3 8B | Mistral AI | $19.35 | Input $0.15 / million tokens; output $0.15 / million tokens; Default rates. | Published pricing | |
| 4 | Mistral Small 4 | Mistral AI | $19.80 | Input $0.15 / million tokens; output $0.60 / million tokens; Default rates. | Published pricing | |
| 5 | Ministral 3 14B | Mistral AI | $25.80 | Input $0.20 / million tokens; output $0.20 / million tokens; Default rates. | Published pricing |
89 products considered; 35 eligible across 7 providers; 54 excluded.
- No eligible active text-model pricing page: 14.
- Unverified pricing: 5.
- Unknown published context capacity: 15.
- Missing token prices: 11.
- Missing source or verification date: 4.
- Non-text output: 5.
OCR pages
Example workload: 10,000 pages/month; published graduated page rates and unconditional allowances; each product's default pricing region or global pricing; no conditional new-customer trials. Product capabilities differ; this compares page charges only.
Selection: Verified page-priced products supported by the OCR calculator, with its freshness, default-region and allowance rules; conditional trials are not applied.
OCR products have different capabilities and page-pricing terms; this is a page-charge comparison. Their official product sources are linked below. Use the OCR cost calculator to change page count and region.
| Rank | Model / product | Provider | Monthly cost | Price inputs used | Official source | Last verified How we verify |
|---|---|---|---|---|---|---|
| 1 | Google Enterprise Document OCR | Google Cloud | $13.50 | 1,000 pages at $0.00 / thousand pages; 9,000 pages at $1.50 / thousand pages; Global pricing; 1,000 pages of unconditional monthly allowance. | Published pricing | |
| 2 | Amazon Textract DetectDocumentText | AWS | $15.00 | 10,000 pages at $1.50 / thousand pages; us-east-1. | Published pricing | |
| 3 | Cohere Parse | Cohere | $15.00 | 10,000 pages at $1.50 / thousand pages; Global pricing. | Published pricing | |
| 4 | Azure Document Intelligence Read | Azure | $15.00 | 10,000 pages at $1.50 / thousand pages; Global pricing. | Published pricing | |
| 5 | Mistral OCR 4.1 | Mistral AI | $40.00 | 10,000 pages at $4.00 / thousand pages; Global pricing. | Published pricing |
5 products considered; 5 eligible across 5 providers; 0 excluded.
All considered products met this use case's eligibility rules.
Use cases we cannot rank yet
Text to speech
2 eligible products from 1 provider; ranking requires at least 3 products from 2 providers.
Proposed workload: 1,000 generations/month; 1 minute of audio and 200 text-input tokens/generation.
Cheapest per token vs cheapest per successful task
A lower list price does not establish a lower cost for a successful outcome. Retries, failed calls, fallback requests and longer outputs can change the total bill. Measure accepted outcomes on your own tasks and include every paid attempt.
Use the retry and fallback cost calculator, read cost per successful task explained, or review the cost-per-successful-task definition.
Related comparisons and planning tools
Cheapest for side projects (1k–100k requests a month) keeps the separate side-project scenarios. Explore head-to-head comparisons, model pricing pages, the AI API cost calculator, or the OCR cost calculator. Prices can change: review the price change log and verification methodology.
Frequently asked questions
What is the cheapest LLM API right now?
High-volume classification: Ministral 3 3B at $5.20 monthly; Customer support chat: Ministral 3 3B at $13.25 monthly; RAG answers: Ministral 3 3B at $8.50 monthly; Coding agent task: Ministral 3 3B at $1.52 monthly; Long-document summarization: Ministral 3 3B at $12.90 monthly; OCR pages: Google Enterprise Document OCR at $13.50 monthly. Each result applies only to its disclosed workload and published list prices; it is not a universal model recommendation.
What is the cheapest LLM API for RAG?
Ministral 3 3B at $8.50 per month for 10,000 requests/month; 8,000 input and 500 output tokens/request; no cache reads assumed.
What is the cheapest API for customer support chat?
Ministral 3 3B at $13.25 per month for 10,000 conversations/month; 6 replies/conversation; 600 system, 90 user and 1,000 retrieved-context tokens/reply; 180 output tokens/reply; 50% history carry. Total per conversation: 12,165 input and 1,080 output tokens; no cache reads assumed.
Is the cheapest LLM API good enough?
This is a price-only ranking. Test models on your own tasks and measure successful outcomes, retries and final spending. The per-task section explains why token prices alone cannot determine cost per successful task. Compare cost per successful task.
How often is this updated?
Rankings are rebuilt from the published pricing dataset on each release. The dataset was last verified 2026-10-21; individual model dates are shown in the tables. Review the price change log for published price changes. How we verify · Price change log.