# BenchLM Data API and MCP

Query current BenchAlign scores and ranks, model catalogs, saved ranking history and cleared benchmark history through REST or MCP. Free includes 1,000 monthly reads.

## Available queries

- Free: current final BenchAlign scores, published ranks, evidence labels, interval bounds, scoring dates and methodology versions for each available surface.

- Free: model identities, source references, sourced release dates and their precision, benchmark definitions, and historical coverage.

- Data Pro: saved BenchLM BenchAlign scores and ranks, evidence labels, methodology versions, and ranking snapshots at a UTC date.

- Data Pro: cleared measurements for an exact model, benchmark and protocol, plus retained benchmark catalog additions, removals and reintroductions.

Current ranking responses contain final BenchAlign outputs, not raw benchmark or upstream ranking scores. Historical dates describe saved repository states. Missing records stay missing. Individual benchmark history contains measurements, not historical leaderboard positions. Check the coverage response before subscribing.

## Plans

- Free: $0, 1,000 successful reads per account per month, current BenchAlign rankings, and up to 10 requests per minute. No payment details required.

- Data Pro: $49 USD/month, 100,000 successful reads per month, the same current data, available historical queries, and up to 60 requests per minute.

- Radar + Data: $59 USD/month, both products billed together with separate API keys and allowances. Link your account from Radar. Existing monthly, annual and legacy terms remain in effect until every current plan ends before bundle checkout opens.

Free allowances renew on the account's activation anniversary. Paid allowances follow the subscription period. Short months use their last day. All keys for an account share usage across REST and MCP; rotating keys does not reset it.

One successful data page uses one read. Initialization, discovery, coverage and usage checks, and unsuccessful queries do not use the monthly allowance. Historical requests contain up to 500 rows and 90 days, with a cursor for the next page. Current ranking and catalog pages contain up to 100 records.

## Connect

Create an account and a key: https://data.benchlm.ai/account

MCP: Streamable HTTP at https://data.benchlm.ai/mcp with Authorization: Bearer YOUR_DATA_KEY. Supported protocol versions: 2026-07-28, 2025-11-25 and 2025-06-18.

Current rankings: GET https://data.benchlm.ai/v1/rankings/current. One model: GET /v1/models/{identifier}/rankings/current.

REST model catalog: GET https://data.benchlm.ai/v1/models. Model detail: GET /v1/models/{identifier}. Benchmarks: GET /v1/benchmarks. Coverage: GET /v1/coverage.

History: GET /v1/history/rank_history, /v1/history/ranking_snapshot, /v1/history/benchmark_history and /v1/history/benchmark_events.

MCP tools: benchlm_list_current_rankings, benchlm_get_current_model_rankings, benchlm_list_models, benchlm_get_model, benchlm_list_benchmarks, benchlm_history_coverage, benchlm_rank_history, benchlm_ranking_snapshot, benchlm_benchmark_history and benchlm_benchmark_events.

Current ranking pages take surface, limit and offset. Resolve exact IDs through the catalog before requesting history. Rank history takes modelKey. Benchmark history takes modelKey, category and benchmarkKey. A ranking snapshot takes at. Use since, through, limit and the returned cursor for historical pages.

Data is a separate product from Radar. Existing website downloads and their terms: https://benchlm.ai/data.

Last updated: October 1, 2026

Canonical page: https://benchlm.ai/data/access