The harness is really important. It matters so much - possibly even more than the model. We had harness crashes after running many agents - granted we were doing quite a bit with it. Grok Build (as a product) review here:
UX friction behind is worse than Claude Code - but seems to be a strange positioning choice - they're more on the 'vibe' side than the 'agentic engineering' things.
Largest issue was actually reviewing output - but if you're going to largely make that opaque from the user, why choose a CLI-based interface that's so mouse-heavy?
There's also problems with the actual model. Thinking is visible, and every interaction goes like this:
"I would like you to investigate adding an API route to tackle x,y,z"
*Grok, thinking: Okay - the user has asked me to add an API route to tackle x,y,z"
Also absolutely absurd other quirks - "I have no tools available in my context" being visible in the CoT.
The auto-approval (yellow, auto-mode) review of Claude Code via Opus is a killer feature - every build-it CLI should be offering this for long horizon tasks.
Messaged one of the engineers about our experience - no feedback.
You'd be better off with Claude Code 5x Max than the 300 USD/month subscription.
Ultimately I've never used Grok and don't want to support the company / CEO behind it, but even without that the 256k context window as given in this link means I won't look at it.
Model intelligence is great, but more often than not I found my issue is not that I would like 5/10% more intelligence, I'm already not even using the highest mode of thinking on most of my queries, and very important queries are better served by asking different model and ask another one to compare and help me decide.
My "big" issues are mostly centered about context loss and how they explode in flight when that happens sometimes without you or them noticing.
I don't know, depends on everyone's usage I guess, and 256k is not THAT shabby, but that's how I feel about it.
I don’t know the methodology at this site, but I noted with interest they rate grok build slightly below Meta’s offering(!)
I would anticipate we will see aggressive improvements / new models at least as the Cursor integration comes together, and a hard push on price and quality over the next few months. Elon is definitely eyeing revenues from the coding LLM market, and has a lot of structural power, since the competitors are running on his hardware right now.
As much as I hate to say this, I'm afraid that Musk is abusing his capital power and Grok Build will end up as the top model eventually.
I tried it on a couple of projects and was (pleasantly?) surprised how good it was. I have no doubt in my mind that spacex needs to be broken up before they grow too big because grok build is going to be the future coding model of choice
They're going to have quite a tall ladder to climb to become the "top model". Currently on LMArena they're nowhere near the top [0] in any of the important categories.
Buying compute that others pay a premium for is not over-buying. Note that $85b GOOG raise is for … building compute they don’t have.
The SX bond issuance reportedly has $90b in demand; they took $25b.
With all of those in mind, it might be worth updating your take on the relative capital power. I’d term it significant. Knowing Musk he will be investing aggressively where he thinks matters over the next few years.
Space-X invests a lot into R&D for Starship which they need to make starlink scale.
They also need it to at least show that in theory they can put a datacenter into space (which is stupid because it would take 300 starship flights to bring up their small colossus dc into space which is only 200MwH).
They also increase the price of Starlink, have to shell out money for cursor.
They overbuyed because it clearly shows that Grok is not frontier or people don't like to use it for other reasons.
Of course Musk has to invest, if what he thinks matters and is susccessful is another story like Cybertruck or his Datacenter idea or his dysonsphere 'idea'.
What a weird thing to suggest. This model is worse than some Chinese models from last year, which were already worse than frontier models from last year.
This looks like a worse Deepseek V4 pro. Costs more, dumber, slower and more verbose than most modern Chinese releases.
Seeing how those Chinese releases are in various states of open source, I would hope a competitor could at least match them.
The harness is really important. It matters so much - possibly even more than the model. We had harness crashes after running many agents - granted we were doing quite a bit with it. Grok Build (as a product) review here:
UX friction behind is worse than Claude Code - but seems to be a strange positioning choice - they're more on the 'vibe' side than the 'agentic engineering' things.
Largest issue was actually reviewing output - but if you're going to largely make that opaque from the user, why choose a CLI-based interface that's so mouse-heavy?
There's also problems with the actual model. Thinking is visible, and every interaction goes like this:
"I would like you to investigate adding an API route to tackle x,y,z" *Grok, thinking: Okay - the user has asked me to add an API route to tackle x,y,z"
Also absolutely absurd other quirks - "I have no tools available in my context" being visible in the CoT.
The auto-approval (yellow, auto-mode) review of Claude Code via Opus is a killer feature - every build-it CLI should be offering this for long horizon tasks.
Messaged one of the engineers about our experience - no feedback.
You'd be better off with Claude Code 5x Max than the 300 USD/month subscription.
Ultimately I've never used Grok and don't want to support the company / CEO behind it, but even without that the 256k context window as given in this link means I won't look at it.
Model intelligence is great, but more often than not I found my issue is not that I would like 5/10% more intelligence, I'm already not even using the highest mode of thinking on most of my queries, and very important queries are better served by asking different model and ask another one to compare and help me decide.
My "big" issues are mostly centered about context loss and how they explode in flight when that happens sometimes without you or them noticing.
I don't know, depends on everyone's usage I guess, and 256k is not THAT shabby, but that's how I feel about it.
I don’t know the methodology at this site, but I noted with interest they rate grok build slightly below Meta’s offering(!)
I would anticipate we will see aggressive improvements / new models at least as the Cursor integration comes together, and a hard push on price and quality over the next few months. Elon is definitely eyeing revenues from the coding LLM market, and has a lot of structural power, since the competitors are running on his hardware right now.
I think this is more of a reflection that chinese labs are full of extremely capable people and it's not something that is trivial to replicate.
As much as I hate to say this, I'm afraid that Musk is abusing his capital power and Grok Build will end up as the top model eventually.
I tried it on a couple of projects and was (pleasantly?) surprised how good it was. I have no doubt in my mind that spacex needs to be broken up before they grow too big because grok build is going to be the future coding model of choice
> grok build is going to be the future coding model of choice
This is a BIG statement, especially considering current benchmark scores, can you expand more, why do you think so?
They're going to have quite a tall ladder to climb to become the "top model". Currently on LMArena they're nowhere near the top [0] in any of the important categories.
[0] https://arena.ai/leaderboard
The coding index is at 51 while the others are at 75
Musk's 'capital power' is spread thinly, as both xAI and SpaceX need ongoing enormous funding and have debts.
Google raised the same amount as SpaceX IPO at the same time, it just didn't get as much attention outside of financial and tech press
https://techcrunch.com/2026/06/03/alphabets-record-breaking-...
The other big players (MS/Goog/Meta) have solid businesses already and both OpenAi and Anthropic are IPO'ing soon.
Musk over bought compute and is renting it out, many of the co-founders of XAI have quit, so doubt they will dominate any time soon.
SpaceX is cashflow positive.
Buying compute that others pay a premium for is not over-buying. Note that $85b GOOG raise is for … building compute they don’t have.
The SX bond issuance reportedly has $90b in demand; they took $25b.
With all of those in mind, it might be worth updating your take on the relative capital power. I’d term it significant. Knowing Musk he will be investing aggressively where he thinks matters over the next few years.
Space-X invests a lot into R&D for Starship which they need to make starlink scale.
They also need it to at least show that in theory they can put a datacenter into space (which is stupid because it would take 300 starship flights to bring up their small colossus dc into space which is only 200MwH).
They also increase the price of Starlink, have to shell out money for cursor.
They overbuyed because it clearly shows that Grok is not frontier or people don't like to use it for other reasons.
Of course Musk has to invest, if what he thinks matters and is susccessful is another story like Cybertruck or his Datacenter idea or his dysonsphere 'idea'.
Musk didn't become the richest man in the world by accident, he is a formidable talent with deep connections.
But no indication that XAI will lead the field any time soon.
I cannot comment on the likelihood of xAI being competitive as an AI lab in the future.
However, their "overbuying" of compute means they can now rent it out for $2.32B/month.
That seems like solid business to me, and raises the question whether your claim of xAI needing 'ongoing enormous funding' is accurate.
Most of that is from Anthropic which is IPO'ing?
Prior to IPO they want to show the best growth numbers possible, so how long they continue with the deal post IPO remains to be seen.
It clearly shows that the billions investedin xAI are not returning what they should with the product they invested in.
What a weird thing to suggest. This model is worse than some Chinese models from last year, which were already worse than frontier models from last year.
This looks like a worse Deepseek V4 pro. Costs more, dumber, slower and more verbose than most modern Chinese releases.
Seeing how those Chinese releases are in various states of open source, I would hope a competitor could at least match them.