Every LLM's price, context window
and capabilities behind one REST API

A total you can check

Ask for the cost of a call and you get a receipt, not a number. Every line names the literal dataset key it was billed with, so the arithmetic can be verified against the JSON.

# 190k input tokens crosses Anthropic's 200k threshold
curl -X POST https://api-llm-specs.axium-lab.com/v1/estimate -H 'content-type: application/json' -d '{
  "model": "claude-sonnet-4-5",
  "usage": {
    "input_tokens": 190000, "output_tokens": 4000,
    "cache_read_tokens": 20000,
    "cache_creation_tokens_by_ttl": { "1h": 8000 }
  }
}'

input.text         190000 x 0.000006   = 1.14    [input_cost_per_token_above_200k_tokens]
output.text          4000 x 0.0000225  = 0.09    [output_cost_per_token_above_200k_tokens]
cache_read.text     20000 x 6e-7       = 0.012   [cache_read_input_token_cost_above_200k_tokens]
cache_write.1h       8000 x 0.000012   = 0.096   [cache_creation_input_token_cost_above_1hr_above_200k_tokens]
                                        ------
                                         1.338 USD

The prompt crossed the threshold, so the whole request was repriced — output included — and the 1 hour cache TTL composed with the long-context axis into that quadruple key. Both decisions come back in resolution, next to the list of every key used. How the estimator resolves rates →

Quick start

git clone https://github.com/axium-lab/llm-specs-api.git
cd llm-specs-api
bun install
bun start          # http://localhost:8080

Or with Docker

docker build -t llm-specs-api .
docker run --rm -p 8080:8080 llm-specs-api

Then ask it something

# The cheapest chat models with a 1M context window
curl 'localhost:8080/v1/models?mode=chat&min_input_tokens=1000000&sort=input_cost_per_token:asc&fields=id,provider,input_cost_per_token&limit=5'

# One model, every attribute the dataset has for it
curl 'localhost:8080/v1/models/bedrock/us.anthropic.claude-3-5-haiku-20241022-v1:0'

What makes it different

Every one of these is a decision taken in the code, not a slogan.

Endpoints

Ten routes. Everything queryable lives under /v1; /health sits outside it on purpose.

Where the data comes from

The catalog is LiteLLM's model_prices_and_context_window.json, MIT licensed, vendored into the repository and redistributed normalized. The prices are the ones LiteLLM publishes — this project does not source them, and it flags the implausible ones rather than correcting them. How the dataset is loaded, and what it costs to be wrong →