Anthropic announced Sonnet 5 this week, described to be as capable as Opus 4.8 but cheaper. Rough timing — Fable 5’s return ate all the attention. I haven’t had the chance to do any coding with it in-depth yet. But from the limited exchange I had over chat, Sonnet 5 still struggled with technical specs. For any technical discussion, I had better luck with Gemini Flash 3.5. In terms of output, Sonnet became incredibly verbose. There is no trace of the laconicism I had noticed in Sonnet 4.

The real talk of the town is Fable 5, and its second trial period, running from July 1st to July 7th. If you were planning to give Fable a go with subscriptions, now is your chance. After July 7th, Fable 5 will be a usage-exclusive model. It appears the community consensus on Anthropic’s subscription models is that they are heavily subsidized compared to their pay-as-you-go counterparts. Some speculate the subscription offers x20 of what API would have charged for the same amount of work. There is technical nuance behind how the math all works out, but the ones actually paying per token are enterprise accounts.

If memory serves, Fable 5 was a landmark model even for an enthusiast at the time of its release. I am aware other models were reasonably close to Fable, but none swept the media with news of suspension. For the lack of a better word, the model was easy to reason with. I didn’t need to rewrite my prompt several times laying out the same conventions and goals. Some described it as tenacious; I suppose its tenacity toward planning a solution made it an easier model to work with. Like I said, I ended up doing more work with less usage on Fable than I did with Opus or Sonnet.

And it also raises an important question about gatekeeping a better model. It’s expensive to be poor — not affording the better model upfront costs more down the line. Not everyone can write off their Fable token usage as a business expense. I must say, for a company that has largely branded itself on AI ethics and public oversight, it is surprisingly lenient on its own business practices. One could have argued Pro and Max are subsidized subscription plans as ‘AI for the masses’, a corporate giveaway, but what does that leave us with Fable paywalled behind usage-exclusivity? Will there be more models like Fable, not available through subscription? Or will the market competition drive Anthropic to open Fable to the masses in the long run?

At any rate, I highly recommend giving Fable a go. With subscriptions, one may use up to 50% of their weekly usage on Fable. Considering the consensus on Fable being token-hungry, the Pro-level plan might not be enough for an actual trial run. But no doubt it would be enough to wrap up a personal project sitting on the back burner. It was that efficient. Whether I am willing to swallow the pill is another question — perhaps one a creative accountant might help with.