Question
Nah usage is low sadly but gpt 5.6 is EXTREMY HIGH
Try the free trial.
ok i will. i did some testing on a website where i can compare fable 5.1 and gpt 6 astra for free and im surprised by how good they are. I literally cant pick lol
Unpopular opinion, don't believe the hype on Astra. I've had issues with instruction drift, confident misinformation being represented and unprompted modifications being made to it's config file to escalate privs. It reminds me of the GPT 5 launch. Capable of so much more than it will actually do. I think they launched too early. The public roll out apologies and free usage resets they're handing out like candy are a lot louder now that I've tried it out.
Lol
I told it to make like a 3d game and in 3 minutes it made a great one. That’s a test I do for LLMs usually. It beat fable in graphics
the limits are so high but you can still get if you use gpt 5.6 sol ultra fast mode but its alot and the resets
Like I used it for all my previous models and just sometimes like 1-2hr wait for 5h but only if you did soo much
My split is:
Claude = my idea to architecture code + other new idsa
Codex: Implementation of everything else
@Banaxi-Tech Oh I see! I’m thinking about making a new SLM called MiniReasoner as G1 and G1 MINI I am postponing for a while lol. What’s a good parameter size sweet spot for fast training and quite powerful?
Sorry for so long, it reccomend the 75-100M or the 40-50M. If you want also 100-140M
Yo orchestrating models is where its at Banana. I throw in Ollama cloud models for reviews or optimizing. I made a reviewer mode the Professor. Adversarial is a role that rewards finding negative feedback a teacher is a role thats obligated to report mistakes but ultimately wants the other model to succeed it reflects well on a teacher who's students are successful i've been getting wicked results with it.