SelfBench builds private coding-agent benchmarks from your own repository’s merged pull requests. Each PR becomes a task with hidden tests, agents and models run against them, and accuracy is plotted against cost with the Pareto frontier drawn in.
Be the first to rate SelfBench. Your feedback helps other users decide.
For each merged PR the task is rebuilt from the commit before the change, with hidden tests and a reference solution.
A task is accepted only if the tests fail without a solution, pass with the original, pass on rerun and survive an independent review.
Public repositories get a live leaderboard on selfbench.dev ranking models, harnesses and reasoning settings.
Similar tools in the same category — compare side-by-side
Abyss is an orchestrator that runs your AI coding agents inside containers. You configure your edito...
agentc is a minimal coding agent written in freestanding C23 with no libc. It ships as a static bina...
Kolo is an open-source alpha that turns a local AI coding agent into a multiplayer session: run agen...
Guides and comparisons about code tools
Five Show HN launches for the accounting layer of agent work: session archives, a Mac home for Claude Code, a usage widget, private benchmarks and a UI critic.
Self-hosted agent platforms, permission kernels, incident agents and simulated environments for teams that need agents inside their own infrastructure.
Ten Hacker News launches that treat agent context, persistent memory and token accounting as first-class problems instead of afterthoughts.
Eleven recent open-source and self-hosted projects for building, sandboxing and deploying coding agents you own outright rather than rent.
Eleven recent projects that turn the Model Context Protocol from a neat demo into deployable infrastructure: registries, compiled servers, and gateways.
