Pinned Tweet
This is how Ordinals Incriptions & BRC-20 makes changes to #Bitcoin . Named Make Bitcoin Great Again.
Don't miss untill the last part
A Thread 🧵
bip341.sats retweeted
ICYMI we've cut the price of Sonnet 5.5 cache reads in half.
In the Claude Platform, it's now $0.10 per million tokens (input is $2, output is $10).
This means Sonnet 5.5 now runs ~20% cheaper on most agentic work.
(API only, no change to Claude Code usage limits.)
bip341.sats retweeted
Announcing dots. Dots work 24/7 for you, learn from your feedback, have their own computer, browser and can be connected to over 4k apps in our ecosystem.
They’re powered by Astra our best model yet. Included in your Pro plan, without drawing down on any of your usage. You can even call a dot while it’s working.
We’re getting you started with your primary dot today and soon you’ll be able to create entire teams of them.
claude code tips:
- use fable only for main session
- all others including subagent must be sonnet
- don't use opus 5 to code!
bip341.sats retweeted
Had both Fable 5.1 and Astra write a project plan. Then had new instances of Fable and Astra compare plans and choose the better plan. Both chose Fable's plan.
The project was to build a benchmark for a workerd feature I'm working on.
Both agreed that Astra's plan was over-engineered. It spec'd out building a load generator from scratch as a node app. Fable checked what I had installed and proposed using vegeta or oha.
Frankly, Astra's plan is also harder to read, over-specifying details with a lot of unnecessary "do not"s.
They used similar token counts but Fable was about 2/3 the cost presumably due to charging much less for cache hits (major pricing change in 5.1 that people didn't seem to notice...).
I also ran a head-to-head writing some C++ code a couple days ago and found Fable's code better-designed and better-documented.
On code review, I had previously concluded Sol was better than Fable, and Astra is definitely better than Sol, able to discover even more, subtler problems.
Of course, none of this is scientific, just my vibes based on a pretty small number of trials.
I'll keep doing more tests but at present I am feeling like the status quo hasn't changed: Claude is better at writing (plans and code), GPT is better at reviewing/debugging.
(And I'm getting really sick of having to re-evaluate this stuff every two months... TBH a "pause" might be nice for that reason alone.)
12hrs non-stop multi agent -> ready for the next button of reset!
most powerfull AI is useless if you couldn't use it as a whole, they don't deserve your money!
1
25















