Kimi K3.1 Pro dropped yesterday. I raced it against Claude Opus 5.5, the top model on Artificial Analysis. Same prompt, a 3D racing game, same 64K tokens.
Kimi shipped a working neon racer in 6 min 40 s.
Opus thought for almost 11 minutes, cost 6x more, and never wrote a line.
Grok 4.7 thought for 22 minutes. Then its racing game drove into a wall.
Mercury 3 built the same game in 24 seconds. It drives. Half a cent.
Gemini 4 Argon took 6 minutes. Black screen.
Same prompt. Round 3 is yours. Reply with a model and I’ll time it.
Same scripted 20 second drive for every game.
Round 1 on the same prompt. Kimi K3.1 Pro 6:40, works. Claude Opus 5.5 spent all 64K tokens thinking and wrote no code.
Grok used 117K tokens. Its provider ignored the 64K cap.
Kimi K3.1 Pro dropped yesterday. I raced it against Claude Opus 5.5, the top model on Artificial Analysis. Same prompt, a 3D racing game, same 64K tokens.
Kimi shipped a working neon racer in 6 min 40 s.
Opus thought for almost 11 minutes, cost 6x more, and never wrote a line.
Both via OpenRouter, reasoning effort high. First run was at 32K tokens and Opus ran out the same way. The gameplay is Kimi’s game as is, I drove it myself.