Benchmark and arena scores per model, with each model's prices in DA.
GET https://opendunes.com/v1/benchmarksFree, with an API key: Authorization: Bearer $OPENDUNES_API_KEY.
curl "https://opendunes.com/v1/benchmarks?task_type=coding&max_results=2" \
-H "Authorization: Bearer $OPENDUNES_API_KEY"{
"data": [
{
"source": "artificial-analysis",
"model_permaslug": "anthropic/claude-fable-5.1-20260831",
"display_name": "Claude Fable 5.1 (Adaptive Reasoning, Max Effort, Default Fallback)",
"intelligence_index": 53.4,
"coding_index": 81.6,
"agentic_index": 57.9,
"pricing": {
"prompt": "0.00156",
"completion": "0.0078"
}
}
]
}| Parameter | Takes |
|---|---|
source | artificial-analysis · design-arena · opendunes |
task_type | coding · intelligence · agentic · search |
benchmark_type | gpqa_diamond · tau_bench_verified_airline · search_browsecomp · search_hle · search_dsqa · search_widesearch |
arena | models · builders · agents (design arena) |
category | an arena category, such as graphicdesign |
search_engine · search_surface | the search benchmarks: the engine, and server-tool or plugin |
include_run_config | include how each run was configured |
max_results | how many rows to return |
A row carries the fields its source publishes: intelligence_index, coding_index, agentic_index for an index; elo, win_rate, tournament_stats, avg_generation_time_ms for an arena; accuracy, primary_score, total_tasks, avg_latency_per_task_ms, avg_cost_per_task for a benchmark run. Some rows carry source: "OpenDunes"; source=opendunes returns only those. Any other source value is a 400 naming the values it takes.
pricing is the model's price in DA per token, as on Models. avg_cost_per_task is in DA.
When the scores cannot be read, the route answers 503 with a Retry-After.