the way chatgpt 6-astra behaves, you guys trained it to behave like a widget! this isn’t the big model; its a flash level model or some fine tuned of something else; it behaves retardedly ; its either you’ll go the tool route or the anthropic route but if you keep staying directly in the middle it will be forgotten; choose a side and stick to it! This model is pretty lame, don’t sell sell a fine tuned flash model at these prices
this astra model is horrendous!
Yeah, there is so much hype going on in the AI world (although I would never deny the astonishing accomplishments it has made at the same time). Don’t trust benchmarks. You gotta try AI yourself on your own projects, trained by your own way of prompting…
I guess you have a Plus account? Astra Work is almost unusable on Plus. I tried it myself. I only got 1-2 prompts. On Pro I can use hundreds of millions of tokens and my last run of Astra Max Thinking (70 minutes) took only 10%. Still quite a lot , but usable … I don’t know what to tell you, other than I feel with you. Don’t get me started on all the things I got disappointed in..
Problem is: All this is so new, I don’t even know how to assess the results I get. Is it efficient? I don’t know. My AI doesn’t know. We don’t have references. We are living in new times… my recommendation is to just wait…
x20 pro
I’ve used these things, they all have their own behaviors sure, but this is a flash level model openai is passing off as the next big thing. Its very stupid in the way flash models tend to be because it’s distilled from something much more capable
this model GPT 6 Astra is Gemini levels stupid!
**No further clarification is needed. This is a partial implementation report—not completion of your requested plan.**
if no further clarification is needed, why do a partial implementation report?
do you know why? it’s a small model used to train, validate, or do synthetic data generation. These are the behaviors of small validator models!
What settings are you using, I am using on either the Codex selected in the ChatGPT app or as a VS Code extension, mostly on Extra High, Fast or not Fast, and it is doing a pretty thorough job for me on just about whatever I throw at it. Better than Fable 5.1.
If these labs do not cross pollinate and think that they can RSI with their own models, King Tutankhamun family tree and short life says that’s a dead end.
After Astra 6 came, work mode is unusable. Before I could work a lot, now it takes all tokens with just 2-3 promts even when using the light version, nonsense! So dissapointed with this.
If your suggestion solution is it works on my machine, that’s not a helpful response…
this Astra model is the scraps of whatever they are using internally; the behavior is consistent with that. I don’t know anyone inside OpenAI but this thing ; GPT 6 Astra performs like one of Gemini flash, or those cheap fine tunes of those open weight models.
They should refund any Astra usage, this thing is terrible, it stops every 3 seconds to ask for clarification; it’s clearly a drafter/verification model that they run in a loop! I
f I wanted a cron job I wouldn’t be paying for compute; this model is one of the worst things that have come from this lab in a long time!