Optimize the serving stack
Each scenario gives you a real workload and its SLOs. Tune the model, quantization, batch, replicas, speculative decoding and cache, and a deterministic model scores you on four axes at once — p95 latency, cost, throughput and quality. No single config wins them all, so the board is a Pareto frontier. Scored on our servers; free to play.
Serve the RAG assistant
Tune a model-serving stack to hit the SLOs at the lowest cost. Four axes, one Pareto board.
Keep the voice agent snappy
A real-time voice agent. p95 latency is king — but hitting it cheaply is the art.
Classify the overnight backlog
A huge offline classification job. Latency is cheap; throughput and cost per request win.
Pass the high-assurance bar
A regulated triage assistant with a 90% quality floor. Quality first, then spend wisely.
The four axes — latency · cost · throughput · quality — are recomputed on our servers from the config you submit, so a forged score is impossible. Only players who choose a handle and opt in appear on each scenario’s public board; we never show your email.