Whether you’re choosing a model for coding, research, or writing—or adding safety checks to an agent—RunLog combines measured benchmark results from nine prepared tasks, token-based cost estimates, and local developer tools.
Start with the task you care about—coding, research, or writing—and adjust how much quality, speed, and cost matter to you. Shortlist candidates from the nine-task benchmark, then estimate your monthly provider bill.
Compare lower-cost models that scored near the top on these nine prepared tasks, model multi-turn token assumptions in the Pricer, and integrate local retry circuit breakers (Loop-Guard) and trace checks (Ship-Check).
Explore task-specific scores, response times, and measured answer costs across nine coding, research, and writing samples.
Estimate token counts and monthly provider charges from your own prompt text, expected output length, and call volume.
Detect repeated tool calls, retry cycles, and observed usage inside your application before loops spiral.
Check recorded agent tool traces from your own test harness against step, tool-use, and cost assertions locally or in CI.
No. The benchmark explorer, cost calculator, and loop simulator run in your browser without an account or payment. Your own provider charges apply if you run models yourself.
Benchmarks measure nine prepared tasks using reference-based AI ratings. Costs reflect those nine completed answers. Use them to build a shortlist, then test your own inputs.
The downloadable Prompt Pricer CLI, Loop-Guard SDKs, and Ship-Check runner are MIT-licensed with public checksums. The hosted website’s full implementation is not published in those repositories.
Built and maintained by Francisco Garrett. Inspect the engineering architecture, test suites, and evaluation provenance on the About & Engineering page →
