MisGraph Hard v1 / Run an evaluation

Evaluate your model

Run MisGraph through your own API or local model. Your browser sends your key only to the endpoint you choose.

Hard v1: 16,384-token cap · 96-request setup check or 960-request full run. Requests are billed by your provider.
01 / Connect your service
02 / Choose what to test

The 12-graph check tests your setup. No automatic retries.

Your run plan

Loading benchmark questions…

Set a spending limit with your provider. Stopping saves partial results; accepted requests may still be billed.

Save your results

Download your settings, answers and scores. Results stay in this browser and downloads never include your key.

Using a local model or another API?

For local models, run a local copy of this site. Other API formats need a gateway that accepts chat-completions requests. Paper-study runner.