←back to thread

Grok Code Fast 1

(x.ai)
504 points Terretta | 6 comments | 29 Aug 25 13:01 UTC | HN request time: 0.739s | source | bottom
Show context
NitpickLawyer ◴[29 Aug 25 16:20 UTC] No.45066063[source]▶
Tested this yesterday with Cline. It's fast, works well with agentic flows, and produces decent code. No idea why this thread is so negative (also got flagged while I was typing this?) but it's a decent model. I'd say it's at or above gpt5-mini level, which is awesome in my book (I've been maining gpt5-mini for a few weeks now, does the job on a budget).

Things I noted:

- It's fast. I tested it in EU tz, so ymmv

- It does agentic in an interesting way. Instead of editing a file whole or in many places, it does many small passes.

- Had a feature take ~110k tokens (parsing html w/ bs4). Still finished the task. Didn't notice any problems at high context.

- When things didn't work first try, it created a new file to test, did all the mocking / testing there, and then once it worked edited the main module file. Nice. GPT5-mini would often times edit working files, and then get confused and fail the task.

All in all, not bad. At the price point it's at, I could see it as a daily driver. Even agentic stuff w/ opus + gpt5 high as planners and this thing as an implementer. It's fast enough that it might be worth setting it up in parallel and basically replicate pass@x from research.

IMO it's good to have options at every level. Having many providers fight for the market is good, it keeps them on their toes, and brings prices down. GPT5-mini is at 2$/MTok, this is at 1.5$/MTok. This is basically "free", in the great scheme of things. I ndon't get the negativity.

replies(10): >>45066728 #>>45067116 #>>45067311 #>>45067436 #>>45067602 #>>45067936 #>>45068543 #>>45068653 #>>45068788 #>>45074597 #
coder543 ◴[29 Aug 25 17:54 UTC] No.45067311[source]▶
Qwen3-Coder-480B hosted by Cerebras is $2/Mtok (both input and output) through OpenRouter.

OpenRouter claims Cerebras is providing at least 2000 tokens per second, which would be around 10x as fast, and the feedback I'm seeing from independent benchmarks indicates that Qwen3-Coder-480B is a better model.

replies(2): >>45067631 #>>45067760 #
stocksinsmocks ◴[29 Aug 25 18:32 UTC] No.45067760[source]▶
There is a national superset of “NIH” bias that I think will impede adoption of Chinese-origin models for the foreseeable future. That’s a shame because by many objective metrics they’re a better value.
replies(1): >>45068189 #
dlachausse ◴[29 Aug 25 19:13 UTC] No.45068189[source]▶
In my case it's not NIH, but rather that I don't trust or wish to support my nation's largest geopolitical adversary.
replies(4): >>45070723 #>>45070873 #>>45071387 #>>45075162 #
1. mft_ ◴[30 Aug 25 00:27 UTC] No.45070873[source]▶
Genuine question: how does downloading an open-weight model (Qwen in this case) and running it either locally or via a third-party service benefit China?
replies(2): >>45071290 #>>45074866 #
2. tipsysquid ◴[30 Aug 25 02:00 UTC] No.45071290[source]▶
from an adversarial / defensive position: the model weights and training data were groomed and known; therefore, the output is potentially predictable. this could be an advantage to the nationstate above the corpo
replies(1): >>45072514 #
3. AnonymousPlanet ◴[30 Aug 25 07:01 UTC] No.45072514[source]▶
This is also true for any US model from a European perspective.
replies(1): >>45074447 #
4. rightbyte ◴[30 Aug 25 13:23 UTC] No.45074447{3}[source]▶
And for any US model from an US perspective. Why is assumed that states are aligned with them self like some sort of CivIII player being coherent and self contained...
5. throw10920 ◴[30 Aug 25 14:14 UTC] No.45074866[source]▶
Genuine answer: the model has been trained by companies that are required by law to censor them to conform to PRC CCP party lines, including rejection of consensus reality such as Tiananmen Square[1].

Yes, the censorship for some topics currently doesn't appear to be any good, but it does exist, will absolutely get better (both harder to subvert and more subtle), and makes the models less trustworthy than those from countries (US, EU, Sweden, whatever) that don't have that same level of state control. (note that I'm not claiming that there's no state control or picking any specific other country)

That's the downside to the user. To loop that back to your question, the upside to China is soft power (the same kind that the US has been flushing away recently). It's pretty similar to TikTok - if you have an extremely popular thing that people spend hours a day on and start to filter their life through, and you can influence it, that's a huge amount of power - even if you don't make any money off of it.

Now, to be fair to the context of your question, there isn't nearly as much soft power you can get from a model that people use primarily for coding - that I'm less concerned about.

[1] https://www.tomsguide.com/ai/i-just-outsmarted-deepseeks-cen...

replies(1): >>45076150 #
6. criley2 ◴[30 Aug 25 17:00 UTC] No.45076150[source]▶
As a counterpoint: Using a foreign model means the for-domestic-consumption censorship will not effect you much. Qwen is happy to talk about MAGA, slavery, the Holocaust, or any other "controversial" western topic.

However, American models (just like Chinese models) are heavily censored according to the society. ChatGPT, Claude, Gemini, are all aggressively censored to meet western expectation.

So in essence, Chinese models should be less censored than western models for western topics.