Qwen4-Max benchmark charts leak: half the tokens, over 60% cost cut

Benchmark charts for Qwen4-Max have leaked, claiming a new MoE architecture that beats Qwen3.8-Max with about 50% fewer tokens and over 60% lower per-task cost, alongside comparisons against Opus5.5, GPT6.1-Sol and Kimi K3.1.

2026-11-16 ~ 2026-11-16 · 2 related posts