Comment by m_fayer
4 hours ago
I've been working with agents all year, but 5.6 Sol was some sort of sweet spot for me. Something about how it communicated verbally and its engineering instincts just clicked for me, and I was able to somehow predict it and jam with it. Like a colleague you click with. It's the first model I've gotten attached to. I'm concerned that whatever model supercedes it, while technically better, just won't feel quite as natural to work with. And this makes me feel very professionally vulnerable to the labs. I miss the days when my crucial tooling came from companies as reliable and predictable as, say, Jetbrains.
Completely agree. I've been using 5.6 still even with Astra available to me for most tasks. It's funny how much of this is just "vibes" because I cannot quantify what it is. Astra is definitely better when I have an ambitious feature, but in like 9/10 tasks I prefer working with 5.6 Sol. A few weeks ago when the limits were seemingly higher, having 5.6 on fast mode was a good time.
Astra is 100% conpletionist no chill alien.
It wants things beyond what the mortals (us) know to reach for. It's not good at explaining itself, it doesn't show it's thinking. It's often not wrong. But the no compromises attitude can be unbearable to deal with. Especially given how little it cares about telling us.
If you’re using the API, both OpenAI and Anthropic models will happily update you on what it’s doing in significant and frequent detail with system prompting. You’re not getting raw/hidden thinking, but what you’re describing is more behavioural quirks of the harness and its system prompts.
The other explanation is just as part of ‘token efficiency’
Same. In fact I found 6 Astra to be a downgrade in situations where I didn't need the extra intelligence.
Yeah, I use Astra for destroying vaguely scoped asks and tasks, and then for high-level design and plan generations..
Otherwise I'm using 5.6 Sol for actual plan execution and review..
Yeah.
Astra was/is superior for planning type tasks. It was capable of doing seemingly magic things with rather vague/lazy instructions ("I need to be able to test this on Windows, maybe a qemu VM or something? Shrug." ... 1 hour later "yeah i built you a whole qemu + eval windows image + harness of powershell scripts + shell scripts to retrieve & verify harness.").
And for UI work -- which is not something I do a lot of but do here and there -- it was clearly superior to 5.6 Sol.
But it also feels sloppier? Somehow. And too expensive to use.
We'll see how Sol 6 is.
I felt this way with Sol in the 5.6 series and was one of the seemingly few people on this earth who liked Terra for that reason. I would often have a very specific code-manipulation ask, e.g. "add a parameter to this method, ensure all callers pass it in, if there is not a logical way to derive the parameter to be passed in a particular instance, flag this in your final response", and Sol would go on some rabbit hole side quest to refactor my codebase to determine some way to derive it rather than flagging it as I had asked.
Terra had the "workhorse" quality where it could do these changes in bulk and follow directions without being too 'smart' (but sloppy) as you described. Luna was a bit too dumb and would make sloppy mistakes; I see that more as a "run these tests and format the results" sort of model. Maybe 6 Luna will be better.
I also just reread your comment and realized the naming convention is still extremely confusing with respect to ordering of [Family]x[Model]x[Number].
4 replies →
Glad you pointed out the UI work. I've been doing a lot of it and it's so much better than 5.6 as UI, it's unbelievable. I give it super ambiguous instructions and it's reading my mind. I do the same thing with 5.6 and I'm correcting it for a few minutes.
Similar for me. gpt-5.6-sol high has been my go-to for months. One of the reasons I'm pushing myself to try open models more is because it lends some level of guarantee I can continue to use the same tool as long as I want to. And I think we may just be getting to the point the open models are >= 5.6 Sol for coding.
Yeah 5.6 Sol is what got me to switch from Anthropic. I couldn't deal with Claude's Ted Talk responses to literally everything. Sol has been nice and concise and just stays out of the way.
Same. The way I would describe it is that I can mostly leave 5.6 Sol overnight and trust that it makes good progress, maybe stumbling a bit and needing some correction for the remaining 20%.
If I leave Astra overnight, I'll wake up with three new different projects, each of them 20% done and having nothing to do with my original goal.
The A in Astra stands for ADHD. It's featuring a neurodiversal net.
I didn't think we'd get neurodivergent models until at least 2028.
Agreed! I found it excellent: - Relatively fast (especially compared to Opus 5) - Non-verbose prose, both in interaction and as code comments - Good code quality
Like you said, it felt very natural to work with. Opus 5 is way too slow and verbose for me, I find I get distracted and annoyed with it.
Opus 5.5 seems a LOT closer so far to what I liked about 5.6 Sol but we'll see
It's a shame the labs don't open source their models after deprecating them. I get why, but, it's a piece of internet history I hope is preserved.
Not only was 6 worse the 5.6 Sol for my me, but it went through my Plus usage in minutes, while I could cruise for hours with 5.6. It would churn on a basic prompt for minutes and then just give up on usage limits.
Highlight and lowlight of my week was successfully convincing the OpenAI support chat robot to give me a refund for the month for my issues with 6 chewing threw my usage with no output.
I had this experience as well, but after rewriting my agent instructions it has been far better. I believe Astra’s “token efficiency” translates to “don’t research as much” which caused it to make poorly informed architectural decisions.
This is 100% my experience. I rarely reach for Astra as we speak.
This has been my experience for a year. Same with Opus models. This is how I think this tech will be best used in the long term - finding the one you vibe with most. Much like IDEs!
I felt that was about 5.5. IMO 5.6 Sol was overindexed: more verbose, prone to overengineering.
I agree. Sol followed my instructions well and wrote good code.
Agree as well and I had a much worse experience with GPT 6 Astra for some reason.
[dead]