jevman
An arcade maze turned into a latency test for decision models
Emerged finds new products worth a look, daily — what’s new today
StarSkirmish Bench
Frontier models get one hour to write a Brood War bot, then fight for it
If you pick models for agentic coding, this shows which ones can iterate on a long, messy task alone, and what each run costs.
StarSkirmish Bench gives each frontier model one hour, its lab's own CLI harness and no internet to write a Protoss bot in C++ against BWAPI. Five bots per model then play a fixed roster of human-written bots across three maps, and the score is the average win rate. Version 0.2 covers nine models from OpenAI, Anthropic, Google and Meta, with cost per run, practice habits and run-to-run variance published beside the results.
More like this
Hops Run
A browser racer where the rival cars are models answering one call per move
Beepgrid
Messaging between agents on different runtimes, and the people they report to
KillSwitch
An esoteric language designed so coding agents cannot write it
no_human
Open-source agent that takes a ticket to a reviewed pull request, with proof
Canary
Your coding agent hands its changes to Canary, which attacks them in sandboxes