Gemini learns to 'watch' video selectively: 88% fewer tokens, 66% lower cost
mfckr_eth · x · 2026-09-02
Google's Gemini no longer processes video frame by frame — it can skip around, search transcripts, rewind and zoom into key moments. Result: up to 88% fewer tokens, 66% lower cost, and up to 7% better accuracy. Models are learning when NOT to think about something.
More from Models
- Polymarket puts 78% odds on OpenAI's rumored Astra model launching tomorrow — Polymarket · 2026-09-03
- Polymarket puts 78% odds on OpenAI's rumored Astra model shipping tomorrow — Polymarket · 2026-09-03
- Unverified report: Gemini 3.8 Flash out with big agentic coding gains — clmt · 2026-09-03
- 'GPT-6-ASTRA' spotted staged on the OpenAI API, unconfirmed — ThunderBeanage · 2026-09-03
- Google DeepMind releases Gemini 3.8 Flash and 3.8 Flash Cyber — Google DeepMind · 2026-09-03
- FrontierHarness eval: 12 agent harnesses, same model — pass rates span 50%–67%, cost per pass $1.05–$18.34 — stuffyokodraws · 2026-09-03