Join the discussion
Write your take first — we'll ask for email only when you're ready to publish.
- Hacker News
- cool
here's Stanford HAI's graph on the carbon emitted from model training per model:
https://spectrum.ieee.org/media-library/chart-showing-estima...
note that Grok's training, thanks to its portable gas generators that are magnitudes less efficient than even other integrated, permanent gas turbines, means the training for this model is dramatically less efficient than models like DeepSeek
a lot of the CO2 emission debate on AI is overblown but it's accurate for Grok
by paimapi - Interesting. Grok 4.5 is a capable model, although not quite at Fable/Sol levels. Will be interesting to see how this holds up. Musk appears to have made a savvy choice buying Cursor's data.by petesergeant
- Nice to see SpaceX on the model frontier! They have been chasing it for a while.by mchusma
- Reading the SWE bickering back and fourth in this thread about Claude vs Grok reminds me of IE vs Netscape bickering way back when.by t1234s
- Netscape 4 Lyf
- /me eating popcorn from my Lynx term
- Grok is not the best model around, but it's decent. It gets the basic job done at a low price. I don't think it can advance frontier Math, yet.by sidcool
- Probably can't advance frontier math yet, yeah. But please let us know other places you want to see Grok improve for future models!by leerob
- It is also not annoying to use. It doesn’t overcomplicate things, and its quick. Much better than eg GLM 5.2.
Pretty good bang for the buck.
- Well this makes me bullish on Gemini if its this easy to reach the frontierby LZ_Khan
- Who said it's easy? xAI staff are putting in 80+ hour weeks and building datacenters faster than anyone.by heaney-555
- I'm just wondering why they sold compute to Anthropic if they were planning on still competing in this race?by rd
- For distillation deals lolby scottyah
- a lot of that compute is used for inference, which is demand-basedby Rover222
- Each minute that a GPU isn't running is money evaporatingby zzzoom
- Timing. They had a massive amount of compute coming online and a serious pipeline of more arriving.
- The revenue was critical to making their IPO numbers look a bit less insane.by Strom
- As a point of comparison: Samsung has sold smartphone chips and later OLED displays to Apple for over fifteen years.
Deals like this that look awkward from the outside but are mutually beneficial to both participants exist everywhere.
by mortenjorck - Competing doesn't mean winning
The rental deal can be terminated by either side with 90 days notice, and presumably Musk would do so if he needed the compute or generally thought it advantageous to do so. For now he doesn't need the compute.
The rental deal may also have been at least in part to juice the SpaceX IPO and to help Anthropic stick it to his enemy OpenAI.
- Likely because they had the capacity to spare. Prior to Grok 4.5, I doubt there was much demand for their models.by JLO64
- Seems the cache read pricing almost doubled from $0.30 in Grok 4.5 to $0.50 in Grok 4.6.
In my experience in heavy coding sessions most pricing is just cache read and cache write like 80% of my token bill.
by pzo - didnt the model 3x in size?by tomr75
- Grok is quite interesting. I run comparisons almost daily on tasks and Grok is its own beast, in a good way.
It's good to have model diversity. When I run a task across Sol, Terra, and Luna, I get variations of the same thing with diminishing quality. It makes the lineup pointless. Ditto for Anthropic. Gemini-3.6-Flash and 3.1 Pro genuinely behave differently. Opus 5 and Fable are.. cousins.
I find that when I want to test a complex creative challenge, having 4 "families" to choose from makes the experience interesting since they will excel in different areas.
Grok might implement unique lighting, Opus, elegant primitives, Sol, accurate snowfall in one pass, Gemini, silky movement. Combined, you can pick and choose best.
For what its worth, Grok always feels "messy" but finishes. Grok 4.6 though is no longer "smart and fast". It's about as fast as Sol though.
A big improvement I noticed in 4.6 was tool use for verification. Previously, Opus/Fable were the only models to consistently screenshot things that they can't directly interact with easily. Now Grok is probably right behind them, perhaps tied with Sol on propensity to verify visually. Grok 4.5 notably did not do this often.
by dudeinhawaii - SpaceXAI is the only frontier model company that had its own compute/date centres and soon chip making factory, I think they will pull ahead with cheaper tokens similar intelligence and better harness/tools. Grok build is 2-5x faster than Claude Code in my opinion.by small_model
- Except EUV lithography is the most complicated industrial process that exists and they won't have usable yields for many years if ever. I don't think Musk actually expects these fans to ever actually make sense they just let him hype and distract.by UltraSane
- > and soon chip making factory
To say that I somewhat doubt this would be an understatement.
by kibwen - >SpaceXAI is the only frontier model company that had its own compute/date centres and soon chip making factory,
Google would like a word. Also, Microsoft.
by asadotzler - >> I think they will pull ahead with cheaper tokens similar intelligence
they just increased cache read from 0.30 to 0.50 - this has the biggest impact on agentic coding. Elon companies have the most expensive everything:
xAI sub: $30 when other starts at $20, pro like sub for $300 where other charge $200.
Expensive electric cars, powerwalls, solar roofs when competetive products/better are cheaper.
by pzo - > I think they will pull ahead with cheaper tokens similar intelligence
They obviously have a huge token cost advantage of the AI labs they are renting compute to, at least for now while they can charge current crazy rates for GPU compute.
by 9cb14c1ec0 - OpenAI is almost there, and Anthropic is pretty close behind. In the next year or two all major AI companies will be vertically integrated to a good degree.by paxys