MiniMax's M3.1 Flash talk: a faster model isn't faster if the job isn't
AI Engineer · youtube · 2026-10-21
At the AI Engineer summit, Morgan Suo of MiniMax used one real research task—an analyst checking whether circuit-board suppliers convert AI demand into cash—to dissect the changes inside M3.1 Flash: sparse attention on long inputs, 4-bit KV cache, low-precision experts, speculative decoding, server load and reasoning budgets. Her core point: a faster model only helps if the finished job gets faster, and every compute-saving tweak can quietly cost accuracy or wait time.
She demonstrated a failure mode where the answer arrives fast but cites the wrong source, then laid out how to judge a model switch: verify sources and calculations, count the full cost of a usable report, and make results repeatable. Slide data shows 1.9x throughput and 7.9s p95 first-token, with aggregate output quality varying notably across configs.
More from Infra
- Data center power set to double by 2030 as grid queues top 2,500 GW — ingliguori · 2026-10-21
- antirez tries DHH's Omarchy on a Dell XPS, plans to adapt DwarfStar SSD model streaming — antirez · 2026-10-21
- NVIDIA reportedly sits on $35B in China orders but can supply only 5-10% — emmanuelvivier · 2026-10-21
- Nvidia's China orders reportedly hit $35B, but supply may cover just 5–10% — emmanuelvivier · 2026-10-21
- willcb: open-sourcing a frontier lab's prod stack is cool, but nothing majorly new today — willcb · 2026-10-21
- swyx: AI compute is supply-constrained — more tokens means more revenue, not lower prices — swyx · 2026-10-21