First proper task with @OpenAI's GPT-5.6.
On the $20 plan, I'm currently using Sol Light for planning and the harder problems, then Terra Medium for implementation.
Still early days, but I'm liking this split so far.
Just added session tracking to Rashun.
I've been doing more contracting work with AI, and wanted to know how much of my usage each project was actually consuming when working out costs.
Now I can start a labelled session, do the work, and see exactly what it cost me in usage.
Just added session tracking to Rashun.
I've been doing more contracting work with AI, and wanted to know how much of my usage each project was actually consuming when working out costs.
Now I can start a labelled session, do the work, and see exactly what it cost me in usage.
Added Cursor to Rashun today.
I don't use it, but with Grok 4.5 hitting opus-tier performance at $2/$6M tokens, it seemed worth adding.
If Cursor's $20 plan included generous Grok 4.5 usage, I'd be tempted. Give me opus-level coding all day over fable-level for 2 hours.
For everyone in NZ waiting on GPT-5.6, yeah it says Thursday, but based on every previous release pattern we're not getting it til Friday 5–7am NZ time
Tried adding ChatGPT to Rashun today (I'm on Plus). Turns out ChatGPT usage isn't exposed anywhere – web API, desktop app, local files all just point back to dead-ends.
Curious how ChatGPT usage is even tracked? I use it quite often and I have never hit limits. Any ideas?
Added a smarter forecasting engine to Rashun last night.
Forecasting makes Rashun more than a usage limit display – it powers usage projections, pacing scores, and warnings.
It adapts to your usage hours and more, helping tell if you'll run out or leave tokens unused.
Shipped a Rashun update today: pacing scores for AI usage limits.
Remaining usage isn't enough – you need to know if your pace gets you to reset.
Now it's glanceable in the menu bar/dropdown via pacing scores and colour cues.
Avoid running out early, or leaving tokens unused.
Been building Dead Kingdoms on the side for the last month. Hardest part so far has been performance with the server running on a Raspberry Pi.
Tonight I added private games so you can host your own world and invite friends via url.
Still rough, but playable. Feedback welcome!
Spent last night talking to people who don't build with AI every day. Not one person mentioned models, agents, or benchmarks.
They were weighing data sovereignty, privacy, trust, and job security.
The gap between what builders and users care about is bigger than I expected.
@brnnnmcdnld@AmpCode GLM 5.2 and Kimi 2.7 have honestly been my go-to open models as well. I've dabbled with OpenCode and Pi but neither has really stuck for me.
Still haven't dared touch 5.5 xhigh😅
Announcing GPT-5.6 on a Saturday morning would’ve been the perfect excuse to spend the weekend building.
Instead, we’re waiting a few weeks.
Hopefully broader access includes those of us outside the US.
What’s everyone’s prediction?
Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work.
https://t.co/OoM83SyISN
@brnnnmcdnld That's a good point. It's definitely pushed me to spend more time with open models recently.
I've particularly been enjoying GLM-5.2 in @AmpCode, though it still has its limits compared to GPT-5.5 in Codex.
What harness and open models you are finding work best right now?