The biggest anti-endorsement of this model is that even a Google employee seemingly trying her best to praise the model is talking about it the same way you would talk about a Toyota Corolla.
I work for Google and am using Gemini 4 Argon. We use it extensively from development to security scanning and troubleshooting, e.g. finding complex networking issues. I am happy and impressed so far.
Google so far has had the ability to produce models with impressive knowledge but terrible real world performance. The best practical model they have is 3.8 flash and that is very mid, often crazy and broken, and expensive. Argon will be at best mid-usable.
There's no way any of you actually believe Google has caught up again after already releasing like 12 models that they claimed to be SOTA that all ended up being utterly unusable. You can't just make amazing models out of the blue. You need iteration.
If GPT-5.6 Sol is removed, there’s no point to stay in Codex.
👎GPT-6 Sol is boring and lacking initiative.
I have absolutely no desire to interact with that model
It kind of worries me that all the suspicion GPT-6 Sol is getting for being cheaper is going to disincentivize labs from releasing cheaper models in the future. I'm still ambivalent about 6 Sol but I've yet to see it do anything that makes me think it's an awful model.
My first GPT-6 verdicts are mixed.
The jury is still out on Sol. I switched back from GPT-6 Astra to GPT-5.6 Sol as my daily driver, and the early vibes/evals haven’t given me confidence to move yet.
GPT-6 Luna is a clearer disappointment.
@pigeon__s 5.6 sol is still better than Astra for me so if 6-sol didn't fix the inconsistency problems that astra has then it's gonna have to be 5.6-sol and opus for me going forward.
🚨 GPT-6 Sol will last 50% longer in our subs compared to 5.6. Luna will last around 40% more.
No one was being clear, so I dug it up myself. We are NOT getting a 2x increase in usage with these new models.
API costs dropped by 2x but our usage has not gone up by 100%.
Told ya guys previously too that API pricing does not reflect subscription usage.
However, this is still good. 50% is quite the meaningful jump.
Linking the official website in comments.
i don't think i've seen many things dumber than this on this entire website. this is up there with patrick bet david challenging netanyahu to a comedy central roast.
.@johnternus acquisition list:
1. @perplexity_ai = search / default LLM
2. @WisprFlow = siri
3. @Superhuman = native AI office suite / mail app / Grammarly is new keyboard on iPhone
4. @exolabs = local open source AI
5. @RobinhoodApp = finance platform (not for sale I know)
6. @Hello_GoAI = enterprise AI
7. @digitalocean = create an AWS competitor
8. Disney = obvious
9. DoorDash and uber = get into the real world of logistics! Bold move!
Spend that @apple war chest on big swings — only takes one breakout
@FriesIlover49 they've been saying they have a model coming soon so we'll see but i'm willing to bet it's not going to be anything crazy. once you get past the facade of it being special cause it's ilya you realize there's really nothing special they can do that a chinese lab can't
@FriesIlover49 as to why i think the company is failing. two years later and all they have is some vagueposting and their ceo and many of their original engineers leaving. low employee count and despite the funding nowhere near the scale for experimenting that a bigger lab has
@mintmarbl I didn't try it on the first day so I can't confirm but I can 100% confirm the ui slop. it keeps generating unstyled html for me lol. haven't seen anything like it in forever.
I'm sorry, Astra is just crazy. BE VERYYY careful telling it to plan anything unless you're line by line reading everything it does. LINE BY LINE. This model will ruin your whole repo and life if you let it do stuff on its own. Be prepared to revert everything you've done.
@mintmarbl Yeah genuinely never using this model again unless it's on very narrow things that i'm verifying meticulously. complete waste of time and tokens and it's much easier to start from scratch than to fix its work. some of the it does stuff is llama 70B-tier.