Founder of a Useless tests and CI directory that runs for way to long
/User/Ai-slop
Joined June 2021
- Tweets3.4K
- Following326
- Followers553
- Likes3.3K
Claude is Great for creating things
ChatGPT is Great for doing things
They compliment each other well, but are incomplete without the other
Did you know you could have Your Dot agent send and read messages in Claude desktop.
So you can let Claude orchestrate Your dot by having it message and read an orchestrator Opus 5.5 task.
Dot for computer use. Opus 5.5 for instruction
Genuinely excited to try T3 code tomorrow. Love Opus 5.5 and already have it delegating to codex but a dedicated Orchestration flow is too tempting to not try out 😎
I don't have access to Claude code projects yet but I set up a workflow that uses one Opus 5.5 High agent as the main Coordinator and several different tracks all working together and its been a dream for the last week.
If Claude Code Projects is anything like this.... 🤯
These are the numbers I was most excited to see!!! Everything else is noise
So I tested GPT-6.1 Sol on real work: 2 repos, 105 planted bugs. Find and fix what you can.
Unlike GPT-6 Sol, which was just a nerfed GPT-5.6 Terra, this one is real. The results:
- GPT-6 Astra (max): 45 for $33
- GPT-6.1 Sol (max): 44 for $6.56
- GPT-5.6 Sol (max): 43.5 for $95.35
- Opus 5.5 (max): 41.7 for $58.53
- GPT-6 Sol (max): 29.3 for $9.33
With a model like this, the 50% cut to the $200 plan doesn't matter.
n=1. More runs and results at more effort levels dropping in this thread over the next few hours 🧵
Sonnet 5.5 is smarter and significantly more expensive per task than Sol,
but not in ways that matter to me.
I’d only use it as a subagent since I don’t use any model on Max.
At its current cost per task, Astra makes more sense as a subagent which is wild to me
On xHigh Sonnet 5.5 more then double as expensive as Astra for a slightly worse score.
great model but not right for my use case
credit to @PawelHuryn bughunt.productcompass.pm/?p…
ive been doing this for 4 days and its been wonderful!!
Claude can now help you build evaluations and hillclimb on them.
In this article, we share guidance on eval design & skills that Claude Code can use to improve your applications.
claude.dev/blog/automating-e…
Even with all the hype the best bang for the money and time is astra For fixing bugs even though technically Sonnet 5.5 will find and fix more.
It just burns a metric TON more tokens/money to do it.
105 replanted, real bugs in two real repos. Blind judges comparing bugs and solutions against the answer key. Always judged by a model from a different family.
More: bughunt.productcompass.pm/?p…
Building local first is the way forward.
Apple foundation models are surprisingly capable and fast.
Combined with Luna medium fallback and you have a capable model that can do a lot in app for free to insanely cheap and fast
Sol 6 and Luna 6 are working well for my use case as sub agents.
It lets me know that I could have made better use of Terra as a sub agent.
Astra Low or Opus 5.5 are great at prompting, evaluating and then adjusting future prompts to get good results from these models
Everyone canceling their 200 dollar OpenAI plans are gonna be pissed when they drop Astra 6.1 next week ( what likely should have been Sol 6)
Opus 5.5 medium is an amazing UI partner
Opus 5.5 High reasoning does great UI and backend orchestration by Sol High and Medium
5.5 Extra is great for Orchestrating multiple agents to debug issues while creating an Eval to find the cheapest model for a specific job in my app
Day 1 Opus 5.5 is the best orchestration model I have used. It is thoughtful, seems to effortlessly just get my intent, has great baseline taste and most of all has yet to make the kinds of stupid mistakes that make Astra feel like im using a Gemini model
I genuinely think that Sol 6 was going to be Terra 6 but they saw that no one was using Terra 5.6 so they decided to launch the updated one along with Luna 6 first. Then they will call what should have been the real "Sol 6" Astra 6.1 latter and say that its a cheaper version.
While everyone else is bugging out over Jev, I just used Foundation models to make easy clarifications on device free.
Apple is a hardware company.
They don’t need to have a great AI. It just needs to be good enough and then route to your personal AI for complex tasks
OS 27 is the first step towards that future.
It’s still absolutely mindboggling to me that Apple isn’t playing in the AI game in any serious way.
It even makes their products noticeably worse because their AI sucks.
Yes, they’ve saved $100 billion by not competing, but they’re losing the war.
Buy someone, do something.