Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
[flagged] Grok is now the most popular model on OpenRouter (openrouter.ai)
58 points by surprisetalk 88 days ago | hide | past | favorite | 46 comments


Misleading because much of Grok's traffic is through their free endpoint. Title should be:

    Grok is the most popular free cloud model on OpenRouter
And even then this wouldn't matter because most devs don't use OpenRouter (commissions!, incompatible API in edge cases, etc.), and most LLM enthusiasts who want to run free models do it on their own machine.

---

Edit: Most → Much


> most of Grok's traffic is through their free endpoint

https://openrouter.ai/x-ai

Code Fast 1: 100B

Grok 4 (free): 84.6B


I was hesitant to use Grok but have actually found it to be excellent and it has totally replaced ChatGPT for me now.

The biggest difference is that it isn’t sycophantic and will often tell me I am wrong when I am. This makes way more of a difference than I thought it would as I feel I can trust the results more (perhaps naively)


FYI, GPT-5 in GitHub Copilot has (almost) completely gotten rid of the “you’re absolutely right” and faux Californian enthusiasm those of us with a British English bias find so condescendingly irritating. ;)

I should try Grok for comparison then.


Haha!

Ok, interesting - will revisit


Hrm, I've never used it, but maybe I should. I've really only used claude and chatgpt, but it's annoying with how they agree with whatever you feed them.


I switched from OpenAI to Gemini a couple of months ago, and was impressed how it sticks to its guns if it thinks I'm wrong.


Interesting, will check it out


I would recommend trialing it for a month instead of ChatGPT.

I still use Claude code for coding


Claude has been super annoying in the last few weeks, I ask it a question and it always immediately starts to write or update code instead of answering.


Same experience here. Codex has been a nice alternative when Claude Code is being dumb


Lol, I've just installed the grok cli and did exactly the same thing..

> I've started to merge scripta and scriptb into scriptc, how would you proceed?

> I've merged the scripts into scriptc

If I have to tell the AI with every prompt not to run ahead it stops being useful..


I don’t mind it because I can enable planning mode and it will (mostly) not modify things.

Claude will sometimes dump very exact file content into a plan.md when I ask it to write its plan down which is annoying.


ChatGPT truly has no moat. I already stopped using it and cancelled my payment. The got left behind so quickly.


As things tend toward commodity (e.g. one item being replaceable with another with not much, or no, difference) the thing that matters most is brand.

So to the contrary, ChatGPT has the ONLY moat so far. 5.8B page visits vs 148m and growing much faster on an absolute basis. https://www.similarweb.com/website/chatgpt.com/vs/claude.ai


The Grok 4 Fast and Grok Code Fast models have really impressed me. The only issue I've had is reaching a rate limit on the Grok 4 Fast model. Amazing pricing for highly capable models with good tool calling support.


They're definitely amazing for the price. I like that you can do quick back and forth with them. but they're not very smart. When I need something to actually analyze or write good code and not just refactor and move things around, they're not good for that.


Agreed. For harder tasks, I like to go to GPT 5 thinking mode, but I'm considering other options.

Some times I've had faster success with some of the larger Qwen3 models (480B and 235B variants). I like them in combination with the Repomix CLI to copy an entire project into context and get a response very quickly with some of the accelerated providers like Cerebras.


Now that Cursor has moved towards a credits system, grok code fast is making the plan last while still being reasonable in inference time. GPT 5 and GPT 5 Codex actually moves my "amount remaining" bar in realtime while being incredibly slow.


Little miss leading as it is sorted by number of output tokens, and I’ve heard Grok is rather verbose.


Sorting by revenue would definitely move Grok down. The top 2 Grok models listed are very cheap or even free right now.


Isn't that a good thing?


Yes, I am a huge fan of the new Grok models!


Maybe it's verbose internally? When I run it, it's just extremely fast. So the verbosity doesn't seem to affect things.


It's very verbose even when prompted to remain brief, but it's still not a bad model and I use it a lot.


Those stacked bar charts are data visualization malpractice. They would tell a much clearer story (who's #1, what are the trends, when did one overtake another) as regular line charts.


50% of their tokens on grok code fast were consumed by kilocode users.

Kilocode has been giving away free grok code fast usage.

When the price comes up for 0 I will be curious to see if this trend holds.


It's free on OpenRouter, but paid on the official xAI API, for the moment.

So, I'd imagine that is inflating the numbers just a bit lol.


Grok 4 fast is a legit model. Their code models, including supernova still aren't smart enough. Claude and Codex are ahead. Its definitely fast, but who cares if you have to re-prompt it or it hits issues it can't fix.


Is open router used a lot? I just use claude code, so this is a misrepresentation of what is actually the most popular model right?


I am using it more than the disaster that is gpt5


I’m curious as to why, since inside GitHub Copilot GPT-5 has been stellar lately. Are you using it directly? (I assume that the prompting strategy inside Copilot is the reason why it’s so good right now).


It’s very slow. Extremely marginal improvement over o3. Sometimes falls short (after taking a lot longer). Lower the thinking amount and all marginal improvement evaporates.


I stopped using openrouter because it’s getting pretty shady. I rather spend my money and traffic on someone like Vercel.


What is shady about it?


Because it's free for now.


[flagged]


For what it is worth, xAI models have historically always underperformed my personal, amateurish benchmarks when compared to what I’d expect to see due to their performance in more publicly known evals. Claude has historically always been the exact opposite, underperforming public evals while outperforming my testing and OAI with 4.1 and 5 output I found to most closely be within what one would expect from those same benchmarks vs my own.

Training to the test, especially in flashy, easy to grasp things like bouncing ball and subsequent failure of similar tasks after small prompts changes made me doubtful of any claims that XAI models are good, with Grok 4 Fast not being an exception. Take the benchmark performance, even when they let external partners like Artificial Analysis do the evaluation, of any LLM lab with massive skepticism, but with XAI I have the most reason to doubt their claims due to prior performance discrepancies.

Just to add some context to what I perceive is their honesty. Also, lest we forget their commitment to publishing system prompts and changes to that for the X deployment of Grok and how these somehow did not reflect any changes during the South Africa and Mechahitler incidents…


This isn't xai, this is openrouter, they sell tokens themselves, so they would know.

Now you should reevaluate the rest of your opinions


Reddit also isn't xAI. The point is you can secretly pull strings to move things around that you cannot directly control.


Fair counterargument. I suspect they probably aren't gaming the OpenRouter stats, but knowing their ethical standards it wouldn't totally shock me.


[flagged]


What are you using it for when it shows a “nazi bias”?

I have used it exclusively for a month and haven’t seen this at all


> The intense Nazi bias makes it a bit of a hard sell to me.

I haven't used it for anything social/political. Do you mean actual 1930s-1940s Nazis or are you using the term analogically?


For what it's worth, barring that week or two where it parroted Nazi talking points, after community notes Grok may be the best thing to happen to the Twitter information ecosystem. The majority of its use in ideological contexts is effective debunking of right-wing conspiracy theories. Trumpists and neo-Nazis constantly argue with it and say it's compromised, and I have seen it sway some conspiracy theorists. You can see many examples here: https://reddit.com/r/grokvsmaga/

I detest Elon and the fact that Twitter is now teeming with neo-Nazis and other nutjobs, but contrary to popular belief Grok has actually been an incredible counterweight to both.


Thanks. I still feel like the social forces around it make it particularly untrustworthy though. I feel very incentivised against using it, and given the wide availability of high quality alternatives, I haven't bumped into any situation where I felt like Grok would be the right choice.

My fallback powerhouse models are Claude 4.1 and GPT5 (and sometimes Gemini 2.5). If those can't do the task I want, trying Grok just seems like a waste of time.


As if name calling was gonna help. It's not just free speech when you agree with it. Competition remains a good thing and honestly, the obvious liberal bias (remember the black female pope pics on Gemini?) on virtually all other platforms makes it an appealing alternative for the more nuanced for exactly that purpose.


Dude, what? "It's not just free speech when you agree with it"? I'm not infringing on their right to free speech by not using their language model. Free speech does not mean free audience. "Competition remains a good thing [...]" agreed, and that's why, like I said in my original comment, I use multiple models (OpenAI, Qwen, Mistral, Llama, Deepseek...). Just not Grok.

I made an X account and paid for premium specifically to use Grok, and cancelled it after a few months of use, because it literally just wasn't useful to me. I don't see why this is a problem? Competition doesn't matter if you don't have the right to not be a customer.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: