8 tasks deleted, 19 rewritten.
Every Terminal-Bench score in your feed this morning was measured on a different benchmark than the one it is being compared against. Terminal-Bench 4.0 shipped yesterday and it is not a sequel, it is maintenance. 2 tasks removed for saturation,
Four models drew their own self portraits on the same sheet and the cheapest one came out looking the best
TypingMind is the only one running this test who put all four side by side instead of posting the winner alone: GLM-5.3-Flash, GLM-5.3, Kimi K3, Claude Opus 5, one prompt,
Claude of Duty is exactly what it sounds like
A tactical FPS in your browser, made by mattshumer_ with one smart prompt, Claude Opus 5 builders, and memory-isolated critics. No art assets. The OG Gauntlet Loop game is now playable on PlayDrop. Link in the first reply.
#pixijs + #threejs handle 200 ships all doing their REAL resource gathering routine.
I use this map to test the performance. Logistics and routing logic done by deepseek_ai , enhanced by claudeai Fable and Opus 5. And you are telling me browser games are not ready for RTS?
You have reached the end of the archive
All of Claude Opus 5