Public evidence. Practical tools.
RunLog helps you explore model results, estimate cost and add checks to your own agent workflow. The public tools do not need an account.
Measured public benchmarks
Compare five prepared tasks with five repetitions per model. Results include quality ratings, token usage, request timing, confirmed costs and failed attempts. Automated judging can make mistakes; these tasks are narrow evidence rather than universal model rankings.
Explore public benchmarks →Prompt cost estimates
Plan input, output, media and conversation-history assumptions with Prompt Pricer. Estimates are distinct from measured responses and provider invoices. Unsupported media and cache assumptions need verification for the chosen model.
Open Prompt Pricer →Local developer tools
Loop-Guard observes repeated actions and usage thresholds inside your runtime. Its website playground is a simulation. Ship-Check verifies signed trace-policy evidence; it does not independently execute your agent or establish answer quality.
Choose a developer tool →Optional sign-in and account previews
You can create a shared RunLog and PA account with Google. PA stays inactive for new users until they explicitly enable it. Private comparisons, Workspace and Teams remain previews under review. Subscriptions and credit top-ups are unavailable. We are improving the public tools before expanding those features.
Existing private, account-owned reports remain protected by account authorization. They are not public benchmark data. Provider processing is separate from account visibility; it is not a guarantee that prompts stay on your device.
Existing-user access →