Last updated: 11/21/2026Last verified: 2026-11-15
Coarena is a public arena and leaderboard for computer-use AI agents from Coasty Systems (YC S26): two agents race the same task on a computer or an Android phone, people vote blind on the winner, and the results feed a Bradley-Terry ranking plus four benchmarks covering capability and safety. It also exposes the same models through an OpenAI- and Anthropic-compatible API with $5 of model usage a day free, and licenses the resulting trajectory data by contract.
What is Coarena?
Coarena is a public arena and leaderboard for computer-use agents, run by Coasty Systems, which its YC company page lists as S26. Two agents attempt the same task on a computer or an Android phone and people vote blind on the result; the votes feed a Bradley-Terry ranking that reports pTop1 and standard error, next to four tracks: Computer Use Arena, Phone Use Arena, CUA KnowledgeBench and CUA SafetyBench. The same models are exposed through an OpenAI- and Anthropic-compatible API with a free daily allowance, and the trajectory data behind the arena is licensed by contract. This listing read the home page, /docs, /leaderboard, /governance, /data and the benchmark methodology page on 2026-11-15, but did not register, vote or call the API.
Key Features
- Blind pairwise battles: two agents run the same task on a computer or an Android phone, and the voter does not see which model produced which result.
- A public Bradley-Terry leaderboard reporting pTop1 and standard error per model, ordered by tokens processed.
- Four tracks: Computer Use Arena, Phone Use Arena, CUA KnowledgeBench and CUA SafetyBench; /mission states 101 public metrics and 24 ranked models.
- Model API for OpenAI SDKs at https://coarena.ai/api/v1 and for Anthropic's SDK at https://coarena.ai/api, plus public /api/v1/models, /api/metrics and /api/leaderboard endpoints.
- The free allowance the /docs states: '$5 a day, and Coarena pays', 30 requests a minute, 2 at once, and a key that needs no card.
- One self-service call to ask for the next step up: $10 a day, or up to 600 requests per minute.
- Trajectory data licensing by contract under /data rather than a public download.
Best For
Pricing
There is no price list: /pricing returns 404. What the /docs does publish is the free tier, '$5 a day, and Coarena pays' with 30 requests a minute, 2 concurrent requests and a key that needs no card, plus a self-service request for the next step up, $10 a day or up to 600 requests a minute. Trajectory data and evaluation work are priced by contract, and this listing found no published rates. Posting tasks, voting and labelling need a Google sign-in; reading the leaderboard does not.
Pros & Cons
Pros
- Human blind voting plus four benchmarks, instead of one headline score.
- Covers both a real computer and an Android phone.
- Publishes pTop1 and standard error, so the firmness of a ranking is visible.
- The same models are reachable through OpenAI and Anthropic SDKs, with a free daily allowance and no card.
- Public models, metrics and leaderboard endpoints for programmatic use.
- Documents its own conflict of interest and keeps the parent company's agent off the board.
Cons
- No published pricing page: /pricing 404s, and data licensing is contract-only.
- Default API access is capped at $5 a day of model usage, 30 requests a minute and 2 concurrent requests.
- Higher limits need a request, and the docs do not say how approval is decided.
- The board moves: it mixes unreleased preview models, and the methodology page says part of the phone comparison uses a model as judge.
- Vendor usage figures for the site are self-reported, and this listing did not call the API or verify them.
Alternatives
Artificial Analysis's open-source benchmark that replays recorded agent trajectories against your own model servers to measure throughput and latency locally.
Generates scenarios from your own agent flow graph and plays them with a simulated user, for testing one agent you built rather than ranking public models.
Browser agents that run your web and mobile app's end-to-end tests, for QA of a product instead of a public leaderboard.
Monitors production agent traffic for hallucinations and schema breaks and ships fixes, for reliability after launch.
FAQ
Details
Platform
Features
- Blind pairwise battles: two agents attempt the same real task on a computer or an Android phone and a person votes on the winner, which feeds a public Bradley-Terry leaderboard with pTop1 and standard error per model
- Four benchmarks — Computer Use Arena, Phone Use Arena, CUA KnowledgeBench and CUA SafetyBench — with published measures for completion, speed, recovery and cost
- Free model API at https://coarena.ai/api/v1 for OpenAI SDKs and https://coarena.ai/api for Anthropic’s: $5 of model usage a day paid by Coarena, 30 requests a minute, 2 concurrent, free key with no card, with self-service requests to step up to $10 a day and up to 600 requests a minute
- Public JSON endpoints for /api/v1/models, /api/metrics and /api/leaderboard, plus a /data programme that licenses computer-use trajectories, human judgments and provenance by contract
Languages
Known limitations
- No published prices: https://coarena.ai/pricing returns 404. The only cost figures on the site are the free API allowance (“$5 of model use a day, and Coarena pays”), the $10-a-day step-up request, and “Terms are scoped per contract” for datasets and evals — there are no plan names, seats, credit packs or dated discount
- A Google account is required for the productive actions; /governance states that “Posting a task, voting, annotating and giving step verdicts require an account”, while watching is open to everyone
- Self-disclosed conflict of interest: the parent Coasty Systems also builds a computer-use agent, and the /governance page says the company “owns the arena and every dataset assembled from it” while keeping its own agent off the roster
- The traffic figures (62.5K people, 6.11M requests, 245B tokens in a 7-day window), the “biggest crowdsourced benchmark” claim and the 24-strong model roster are the vendor’s own numbers; the roster includes unreleased preview models, and the Phone board has comparisons judged by a model rather than a person, as /benchmark/methodology discloses. Third-party bots exist to farm the free tier, and coarena.com is an unrelated for-sale domain


