Multi-token prediction is bolted onto a frozen on-device model

Claim

Google bolted Multi-Token Prediction onto frozen Gemini Nano on Pixels for more than 50% faster inference with bit-identical output, while Alibaba's Wan Streamer listens, sees, thinks and answers on video in real time at 25 frames per second.

Reported metric: +50% speed, 25 fps

Reporting date: 2026-06-27 (not necessarily the event date)

Sources

Attribution and reporting linked by this record; links do not imply independent verification.

• Welcome to June 27, 2026 (theinnermostloop.substack.com)