| name | tokenomics |
| description | Get fresh, live LLM pricing and cost estimates via the tokenomics CLI or MCP server. |
tokenomics — live LLM pricing for agents
Prices come live from the models.dev catalog (thousands of models across 140+
providers). There is no local database and no cache to refresh — every call
reflects current data.
Units (read this first)
- All token prices are USD per 1,000,000 tokens. A value of
2.5 means $2.50 per 1M tokens.
estimate/estimate_cost take raw token counts (e.g. 1000000), not millions.
output_per_mtok: null means the model is non-generative (embeddings/rerankers); output tokens cost $0.
Rules
- Use
--output json (or rely on the default: JSON whenever output is piped) and parse it.
- Add
--fields a,b,c to every search to keep responses small (masks the rows under models).
- Resolve real model ids with
search before get/estimate. Model ids are provider/model, e.g. openai/gpt-4o.
- For "which is cheaper for my workload", use
compare / compare_models — it does the math; don't compute prices yourself.
- Model and provider ids are plain identifiers: no
?, #, %, /-prefixed paths, or query params.
- On
MODEL_NOT_FOUND, read details.suggestions or run search; do not invent ids.
CLI quickstart
tokenomics search "gpt" --max-input 1 --fields model_id,pricing --output json
tokenomics get openai/gpt-4o
tokenomics estimate openai/gpt-4o --input-tokens 1000000 --output-tokens 200000 --requests 10
tokenomics compare --models openai/gpt-4o,google/gemini-2.5-flash --input-tokens 1000000 --output-tokens 500000
tokenomics providers
tokenomics schema estimate
MCP tools
search_models, get_model_pricing, estimate_cost, compare_models — all read-only.
Each returns JSON with fetched_at. No write/refresh tool exists because data is always live.