Terminal Bench

Real agentic task execution — groundtruth → local model → cloud. No faking it.

Groundtruth Local 30B Cloud Escalation tmux bash

Tasks Solved / Total

—

No run yet

Total Spend

$0.00

Local-first: mostly free

Mean Turn Latency

— s

avg per task

Live Run

No active run

No active benchmark run

A smoke test runs 10 representative tasks across all categories. Estimated cost: $0.20 – $0.80.

#	Task	Category	Turns	Source	Cost	Solved	Time

No benchmark run yet

muLLM's agentic loop hasn't been benchmarked yet. Start a smoke test to run 10 tasks across filesystem, git, scripting, and debugging categories.

Estimated cost: $0.20 – $0.80 — most tasks hit local or groundtruth ($0)

—

No data — run a benchmark to see results

—%

Groundtruth hits —

Local solves —

Cloud assists —

Failed —

Total spend: $0.00 Cloud only: $0.00 Groundtruth + local: $0.00