Google I/O Digest (2024 + 2025)

🇨🇳 中文版

2024 → 2025: From “Announce” to “Prove”

Across 15 keynote videos, the corpus reveals a sharp shift in Google’s I/O messaging strategy:

Metric I/O 2024 (9 keynotes, 3.7h) I/O 2025 (6 keynotes, 3.7h) Δ
Products announced 89 57 −36%
Benchmark numbers on screen 6 16 +167%
Demos 37 31 −16%
Forward time commitments 69 45 −35%
On-screen text moments 89 65 −27%

The 2024 main keynote announced 36 products with zero on-screen benchmarks. The 2025 main keynote announced 16 products with 7 benchmark numbers on screen (LMArena Elo 1448, WebDev Arena +142, USAMO 2025 49.4%, LiveCodeBench v6 80.4%, …).

The narrative shifted from “look at all the new things” to “look at how much better the existing things are.” Forward commitments also dropped 35% — fewer promises about what’s coming, more evidence about what’s already shipped.

What’s in each keynote page

  • Segment-by-segment breakdown with clickable YouTube timestamps
  • Every product announcement (name, version/status, features, pricing/availability)
  • Every benchmark number shown on screen (with comparison context)
  • Demos (with success/failure flag)
  • Forward time commitments (tracked, for verifying delivery rate later)
  • Notable quotes (Chinese version adds translation)
  • Visual signals: on-screen text moments, stage moments, visual demos, production signals

Browse

Future work

  • Commitment delivery rate: I/O 2024 made 69 timestamped forward commitments. How many of those were actually shipped by I/O 2025? This is a falsifiable research question the corpus enables.
  • Multi-year narrative arc: extending to 2019-2023 would let us trace Google’s AI messaging evolution (LaMDA → Bard → Gemini → Gemini 2.5).
  • Benchmark citation network: which I/O sessions lean on benchmarks vs raw demos? Useful confidence signal.

License

CC BY 4.0 for the indexed notes. Original keynote videos remain the property of Google.

By @QihongRuan · 2026