Can someone explain to me why Gemini 2.5 Pro coding agent is so bad compared to Claude when Google is supposed to be you know... Google. They literally have infinite funds like Microsoft, and WTF is even Anthropic? I've never heard of them but suddenly their coding agent makes magic
>>107082806they've been around for years now, wtf are you smokingand how exactly are you quantifying how much better claude is than gemini?
anthropic was founded by people who worked at openai but thought they weren't taking ai safety seriously enoughthey've been /aicg/'s choice of model for cooming for years now
>>107082806sir just wait for gemini pro 3
>>107082806Gemini 2.5 Pro has been gorilla tier single digit IQ recently.It desperately needs a major upgrade.
>>107082806yandex worked for finding the source
>>107082919>they've been /aicg/'s choice of model for cooming for years nowIt'll never be not funny how Sonnet 3 was the most horny LLM ever created, including for local and "uncesored" LLM
>>107082887Have you even used Gemini 2.5 Pro coding agent? It's absolute garbage and I've read that as well from anyone, I'm not talking about ChatGPT level of garbage but much worse, it's really really bad
Google is a curry sweatshop
How do you make Gemini make things in another language?It only codes web pages
>>107083366I've been testing it out for the past few days but since I'm not a habitual vibe coder it's hard to judge whether the results are better or worse than what other models offer. Only using it because I already had a google billing account set up for other services.The project I've given it is a fairly simple website and it seems mostly able to handle that, but I'd need to try another agent to compare.Which one is the best currently? How exactly does claude or chatgpt or whatever differ? What exactly are the pain points that you notice?
>>107083930How do you turn a game made in Gemini into an Android apk file?In the preview window my game works. But Gemini won't convert the game into anything useful.
>>107083973The absolute state of vibe-niggers lol
>>107083973well I guess if that's the level of instruction you're giving it I can see why it'd fail lmaoI've been watching what it does and providing it the sort of instructions I'd give my junior devs. Am I supposed to infer that with something like claude I can just tell it to make a whole ass project for me and it'll succeed?
>>107084124There was a lot of back and forth to fix bugs though
>>107084248what's a workflow that works for you in claude? do you ask it to generate detailed plans and do step by step? do you check what it's doing and make suggestions? do you use tricks like multiple agents?or is it literally just>build me an app that does X and then leaving it on auto accept?
>>107082806ChatGPT just werks
>>107082806Thank you for sharing that image, I knew it must have been a crop and was not disappointed when I found it.
>>107084124No you still need to review Claude's homework.
>>107082919I do cunny with 2.5 pro and it is nice, just prompt it correctly.
>>107082806for me, it's deepseekI would rate deepseek > gemini > chatgptdidn't try the other ones though
>>107083973you should copy paste whatever you are doing in android studio and click build > generate signed apk
>>107084839Kimi is also good it's free on Nvidia Nim, it also does cunny very well
>>107082806Something about Claude's initial planning steps make it better at executing code. That or Gemini has shitty default settings I never bothered with. But I also noticed that Gemini has far more trouble fixing issues and adding components to existing scripts. I can give Claude and Gemini both explicit instructions on what I want added to a script and Claude will have it done in 1-2 prompts while Gemini will still utterly fail to do anything significant 4 prompts later.
>>107083398And the sky is blue.Are you going to tell us next that the oceans are very wet?
>>107084542>uninspired coomerbaitget some taste, pleb
>>107084319I told it to straight up make what I needed. It actually made something better than what I wanted but there were bugs when I tested it so I can talk to it for 30 minutes to fix it. I tried to deploy it locally but it wasn't easy.>>107084839I can't find deepseeks coding environment, it's the best at trading stocks so it's probably also the best at coding.>>107084870Ok but is there an environment like in Google ai studio?
>>107082806One of the things anthropic does that others don’t, is buy every single physical book, every unique ISBN they can get ahold of, and use a machine to rip them up and scan them.So Claude has every programming and system design textbook ever written in its training set. Google doesn’t.
>>107083930I'm the one you replied to, I've used it for both front end work and backend work, when it came to front end it made an absolute mess trying to change just little things, like it changed entire unrelated things to the task. Same with backend, it basically fucked and started touching where it didn't have to touch, or made just a poor solution.Currently, the best agents you can find like the best is Claude Code 4.5, I have acess through Anthropic API and I use it through their VSCode extension. It just works, like I'm not even exaggerating it's insane, but very expensive. This is followed (in my experience) by GPT-5 (On high) through their official extension Codex. It is much cheaper, I've been using their basic Pro plan and it works enough for backend, except for frontend work which kinda sucks.Then I've used Github Copilot, where you pay $10 and have access to a lot of models, including Claude 4.5 and GPT-5, but I think they're a little dumber than using the direct API access, I mean they must be, but still for front-end work I use this Claude 4.5. The ACTUAL full battery Claude I have to save it for hard backend work because again, it really, really fucking works.
