We built a custom inference engine to deliver fast and cheap inference. We optimized this engine for coding and research workload and achieving an average speed of 469 tok/s and 4.5x lower costs than Openrouter. Software development at that speed feels different. Get an API key and try in your Opencode!