Added GLM-5.2, MiMo v2.5 and 3 more models
Five more models joined the matrix this week, and we re-ran every prompt against them. The alien-shooter build is a good stress test: it needs a game loop, collision, and input handling all in one file, so the gap between models shows up fast.
DeepSeek-V4-Pro shipped a playable loop on the first try, while several of the cheaper models stalled on a blank canvas or a control panel that never wired up. Open any tile below to see the live build and the full transcript.
Top builds in this update
OUTPUT deepseek-v4-pro Top-down alien shooter
Generated pi-agent · 196.2s · $0.0170 · 9.6k out
Full run + transcript →
OUTPUT kimi-k2.7-code Top-down alien shooter
Generated pi-agent · 699.9s · $0.1491 · 20k out
Full run + transcript →
OUTPUT glm-5.2 Top-down alien shooter
Generated pi-agent · 476.7s · $0.2385 · 25k out
Full run + transcript →
OUTPUT minimax-m2.7 Top-down alien shooter
Generated pi-agent · 146.6s · $0.0098 · 6.2k out
Full run + transcript →