I’m freaking out about Sonnet 5
In a Nutshell
Anthropic released Claude Sonnet 5, which is only marginally better than Sonnet 4 but costs 35% more due to a new tokenizer that uses more tokens. Export controls on Claude, Fable, and Mythos have been lifted, ending the subsidized $200 subscription access that provided massive Fable usage. Fable now requires paid API access at high cost, while the speaker pushes for maximum automation of all tasks.
These notes were generated by AI and may contain inaccuracies.
Anthropic has announced Claude Sonnet 5. The model is not better than Opus or Fable. Sonnet 5 is better than the previous Sonnet model, described as Sonnet 4 + 1. The performance difference is described as not close when examined closely.
The new model includes a new tokenizer that uses up to 35% more tokens. This increase in token usage makes the model more expensive to operate. To ease the transition, Anthropic is lowering the price for a couple of weeks. After this promotional period, pricing returns to normal Sonnet rates while still using 35% more tokens due to the tokenizer change.
Anthropic announced that the US government has lifted export controls on Claude, Fable, and Mythos. Access restoration begins today. The speaker notes this as the end of a prolonged period without access. The grace period for accessing Fable through the $200 Claude subscription, which allegedly provides $6 billion worth of usage, has ended.
Fable must now be accessed through the API. The API cost is described as equivalent to three engineers, ten European engineers, or 6,000 Indian engineers. The speaker expresses enthusiasm about the return of AI progress and the ability to automate tasks extensively. The speaker advocates for automating everything possible, including tasks already automated, and states that employment has only provided depression and commutes.
Sign in to read the full notes
Get access to AI-generated notes, topic timestamps, and more.