Google gets ready for the tokenmaxxing hangover
Notes from Google I/O, including what Sundar Pichai told me about Mythos. Also: What Andrej Karpathy is doing at Anthropic, Microsoft's special CEO summit guest, and partnering with Parallel.
I spent Tuesday in Mountain View at Google I/O watching the keynote and speaking with leaders from Google both on and off the record. Below, I have my initial takeaways. I’ll have more to share in the coming days.
I reported last week that Google’s next model wouldn’t push the frontier and instead land in the ballpark of GPT-5.5 and Claude Opus 4.7. That’s what Gemini 3.5 Flash, released across Google’s platforms on Tuesday at I/O, turned out to be.
Across my conversations with senior executives, I think Google is positioning for a moment with Flash when companies look at their token bills, ask what they’ve been buying, and look for more efficient altneratives.
The tokenmaxxing hangover
During the main I/O keynote, CEO Sundar Pichai correctly noted that many companies have burned through their annual token budgets (it’s May) and framed Gemini 3.5 Flash as the off-ramp. A trillion-tokens-a-day customer could save $1 billion-plus annually by moving 80% of its workloads to 3.5 Flash, he said.
‘Tokenmaxxing’ has become a meme in Silicon Valley, driven by OpenAI and Anthropic’s API economics, and people spending heavily on tokens are starting to realize that much of the spend is wasteful. Google clearly sees the hangover coming.
