Free
$0/ month
1,000 successful reads per account each month, 10 requests per minute, current final BenchAlign rankings, licensed benchmark results, and catalogs. No card required.
Start with FreeContinue to your Data account
Checking your account link…
Radar and Data keep separate plans, API keys and allowances. This link signs you in to Data.
Your Data account
Connect your tools to BenchLM.
One API key works across REST and MCP. Start with current BenchAlign rankings, then add recorded history when you need it.
Data Pro
$49USD / month
100,000 successful reads per month, 60 requests per minute, and available history from the last three rolling calendar months, measured by UTC date.
Sign in for ProResearch
$249USD / month
500,000 successful reads per month, 60 requests per minute, and all retained eligible history with no age cap.
Sign in for ResearchAllowances are shared across REST and MCP. Free resets on the monthly anniversary of activation; Pro and Research follow the subscription period. History has gaps by source, model, and date; check coverage before subscribing.
Email sign-in
Open your account
You chose Data Pro at $49 USD/month. Sign in to review the plan before upgrading.
Enter your email to receive a one-time code. New accounts begin on Free.
Your workspace
Data access
Monthly allowance
— reads left
API keys
All your keys share one account allowance. You can keep up to five active keys.
Save this key now. It will not be shown again.
| Key ID | Created | Action |
|---|
Create a key to connect your first client.
Connect a client
Use a Bearer token with your API key. Keep keys in your client’s secret settings.
- MCP endpoint
- REST base URL
- Authorization header
Bearer YOUR_DATA_KEY
MCP clients should use the Streamable HTTP transport. For REST, start with this catalog query:
Your plan
Data Pro $49 USD / month
100,000 successful reads per month, 60 requests per minute, and three rolling calendar months of available history by UTC date. Check coverage before choosing a plan.
Not available for purchase yet.
Research $249 USD / month
500,000 successful reads per month, 60 requests per minute, and all retained eligible history with no age cap, including records older than 12 months where available. Gaps remain by source, model, and date.
Not available for purchase yet.
Radar + Data $59 USD / month
Radar and Data Pro’s three-month history window, with separate API keys and allowances, billed together. Existing subscriptions keep their terms until period end before bundle checkout opens.
Bundle checkout is being prepared.
Link your account from RadarQuery reference
What you can query
Free access includes current final BenchAlign scores and ranks, licensed benchmark results, model identities, and benchmark catalog metadata. Pro includes three rolling calendar months by UTC date. Research includes all retained eligible history with no age cap. Both require available coverage for the requested fields.
| Data | REST endpoint | MCP tool |
|---|---|---|
| Account usage and resets | GET /v1/usage | benchlm_get_usage |
| Current BenchAlign rankings | GET /v1/rankings/current | benchlm_list_current_rankings |
| One model’s current rankings | GET /v1/models/{identifier}/rankings/current | benchlm_get_current_model_rankings |
| Model catalog | GET /v1/models | benchlm_list_models |
| One model | GET /v1/models/{identifier} | benchlm_get_model |
| Benchmark catalog | GET /v1/benchmarks | benchlm_list_benchmarks |
| Licensed benchmark result coverage | GET /v1/benchmarks/results/coverage | benchlm_benchmark_result_coverage |
| Current licensed benchmark results | GET /v1/benchmarks/results/current | benchlm_list_current_benchmark_results |
| Published evaluation runs (Pro / Research) | GET /v1/history/benchmark_results | benchlm_benchmark_result_history |
| Publisher result snapshots (Pro / Research) | GET /v1/history/benchmark_result_snapshots | benchlm_benchmark_result_snapshot_history |
| Available historical coverage | GET /v1/coverage | benchlm_history_coverage |
| Stored BenchAlign rank history | GET /v1/history/rank_history | benchlm_rank_history |
| Benchmark measurements over time | GET /v1/history/benchmark_history | benchlm_benchmark_history |
| Benchmark additions and changes | GET /v1/history/benchmark_events | benchlm_benchmark_events |
| Stored ranking at a date | GET /v1/history/ranking_snapshot | benchlm_ranking_snapshot |
Sign in to check available historical methods and dates.
Current rankings take surface, limit and offset. Use catalog IDs to choose a model and benchmark. Ranking history takes modelKey; benchmark measurements take modelKey, category and benchmarkKey. Add since, through and limit to choose a historical page. A ranking snapshot takes at.
Historical queries return up to 500 rows per page. Saved-state history covers up to 90 days per request and uses committed record dates. Source-run history preserves the publisher’s reported run start; use date filters of up to 90 days. ECI release dates do not establish evaluation run times. BullshitBench v2 contains one pinned snapshot with native judge ratings, rates, counts, and published ranks; its artifact generation time is not a per-model run date. Arena overall snapshots preserve the publisher’s rating, rank, confidence interval, variance, and vote count, with three scoring methods kept separate. Publication dates do not establish model run times. A ranking snapshot contains the covered subset and identifies its coverage.
Current ranking responses contain final BenchAlign outputs and their scoring version. Licensed benchmark result responses contain source scores with units, source model profiles, and attribution; filter them by category or boardId. For source-run and publisher snapshot history, use boardId and sourceModelId from the result coverage and current-result tools. Snapshot history accepts since, through, limit, and cursor; optional artifactId selects an exact retained artifact. Historical rankings are stored BenchAlign outputs. Benchmark measurement history contains scores. Published Arena snapshot ranks come from the source’s dated leaderboard; run records do not establish ranks.
Each successful data page uses one read. MCP discovery, coverage and usage checks, and unsuccessful queries do not use your monthly allowance. Free allows 10 requests per minute; Pro and Research allow 60. Use GET /v1/usage or benchlm_get_usage to check both reset times. API key rotation does not reset usage.