Head-to-heads

Model vs model, same prompts

Two or three models on the same real prompts, judged blind in both orders. Wins, cost and latency side by side.