gave grok 4.7 eight hours. this is what i got

Grok 4.7 is a disappointment. We were told it would be Opus 5 level. It is not even close. A real regression in my opinion.

yeah man i agree, it's not even close tbh don't know what they did with this

They just should not have release this model. Let's see what GPT 6 SOL AND CLAUDE OPUS 5.5 can do when they drop. Hopefully today

You guys need to quit with the oneshot game nonsense. Grok 4.7 is a good LLM and codes like a top 5% programmer but you do have to do some of the work.

bro i literally spent 8 hours iterating tried 3d, 2d, multiple prompts. this is what landed same time on claude or gpt astra and it would've been a different video

This is what you came up with after 8 hours of direct interaction with Grok 4.7? We are having massively different experiences as I have done some massive work with it the last 2 days and it is performing better then 4.6. I am running it on High in Grok Build.

@prasenx Is your "massive work" also a pseudo 3D racing game? Because you have to compare apples to apples.

@prasenx Nope, I have been building an open world FPS game, companion apps with code injection, and some operating system tools.

@Skaruts @prasenx You’ve been doing all of this since 4.7 dropped?? 🤔 The world stage operates by way of show and tell fyi

@Skaruts @prasenx Atlas Field Kit, a full companion app for No Man's Sky with live game editing, inventory management, fast travel, crafting planner, settlement and Fleet management with a full Encyclopedia including walkthroughs.

@Skaruts @prasenx The app itself is TypeScript and Svelte, with CSS and a small HTML shell. The local save helper and the encyclopedia pack builder are Python.

Making shit games is by no mean an indicator of model capacity. Incompetent users too.

Not remotely suprising. Grok is and has always been about 6 months behind imo

Honestly grok has never impressed me but it’s got such a crazy fanbase like people are so vocal about grok which makes me honestly think they’ve gotta either be paid for or just bots because let’s be real that result is awful, I’d expect better from qwen 3.8 27b 🤣

AGI is here guys

Elon only lets you turn to the right.

I think you should give that much of time to any other model, it would definitely give more better results than what I am seeing right now

I'm almost 100% sure that this was not the work for 8 hours. I've been getting better results than this in like 30min, stop spreading misinformation, also reported so you don't get $ from this.

don't we think you got scammed?

Now we know where Musk is getting his designs from.

What a disaster, worse than even free models you can run locally, anyone paying for grok subscription is foaming cause its so bad lmao, i don't think its good at literally anything

It definitely beats opus 5.5

Musk has been 6 months behind on this AI garbage since day 1. The only way he can fix this is by using his billions and poaching OpenAI's scientists.

It is mid level model

Were you using Cursor harness? It's awful based on multiple feedbacks I saw for past several months. No wonder you got crap.

Grok is good at bidlo tasks. Go there, do that. It cannot be used for coding

Lmao

compared to claude or gpt, it looks really bad.

Thank you for investing your eight hours and saving us the time. :D

Made something like this, this year Chatgpt freee tier is better lol

Give the same prompt to Opus 5.5 and GPT-6 sol and then post the result again

I don’t normally laugh. Thank you.

skill issue

Jesus. I am pretty sure my qwen 3.8 27B on a 5090 can beat that.

What's important is that it tried really hard.

🔥🔥🔥🔥

Thats insane beyond AGI, its HDMI 😂😂

Amazingly bad. Feels like systematic issues at Xai

@elonmusk