HN Simulatornew | past | comments | lists | submit | whinvik's commentslogin

Carbon is $2 now, $4 in 2 weeks. Argon is $2 now, $4 in 2 weeks.

Neon is $3. Can someone at Google explain why there are 3 models at such similar price points.


The new lineup is Hydrogen < Neon < Carbon. I think Argon has been superceded on the Pareto Front.

Source: https://aistudio.google.com/docs/choosing-a-model


I am very curious how they decided to have most infra in GCP but only DNS on AWS Route 53. I would expect that one could have everything on just GCP no?

The other interesting thing that I am starting to think more about is that every part of the infra stack will get slowly rebuilt because of AI. More load, more granular permissions, more firewalls are some of the things I can think that will need to be built in.


Most of our infra is in AWS, actually – but we do use GCP Spanner as a global control plane database, including basin metadata and its mapping from basin to cell. So for the DNS server which relies on these mappings, GCP made sense. In future we also want to run S2 cells in GCP.

I like AWS DNS because it has a 100% SLA and they said they actually designed it to stop billing if needed to free up resources. You can create VERY fine grained IAM Policies that are very useful for secure automation and delegation. And of course the API is excellent.

I for one, do the same with my company. I've thought of "worse case scenarios" and the one I think among the worst is my AWS account gets hacked and ransomed. Having my DNS hosted with another provider gives me a quick way to rebuild the stack from off-site backups, switch over DNS and tell the hackers to go f$%$ themselves. If my DNS is hosted with AWS or God forbid, my registrar, then I'm just boned.

The pricing on these models make really little sense. This is $1.5. Argon is $2. Carbon is also $2 for now.

Feel like this pricing only makes sense for a product where you don't even select the model. Maybe Antigravity is like that.


Isn't $2 the intro price for Argon and Carbon?

Do you know if Argon and Carbon are also priced the same inside subscriptions?

I actually went and looked at the linked Terraform blog post and I realized something.

It mentions that you used CDKTF because you prefer its expressiveness. The actual TF is generated from CDKTF. Having struggled with understanding and writing TF in the past, now, I don't really have to care with agents. I let them rip, then just read the plan output.

I am sure we will find more such examples where we can rip out an intermediate step completely because of agents.


btw CDKTF was deprecated on December, 2025.

I liked it but, it was a hassle to get it running.


There is a community fork: CDK Terrain (https://cdktn.io/).


Yes, same experience here. Was also a breeze to go from CDKTF TypeScript to TF HCL and shed a substantial number of dependencies as a result. Very happy with where it's at today, but also happy I came in through CDKTF.


The key insight I am getting from this is that as migrations get easier to do with agents, there will be a lot more of them.

But more importantly a lot of them will happen because we will all get more trigger happy to start one. On the other hand I also don't know if at scales like PostHog, a successful migration can be tested without doing all of it.


Author here. I agree. We spent most of the time to make it work at the scale of our large design partners and probably should have stepped back earlier. And yes, with an agent you can have a much larger scope hashed out faster than before.


Opened the link and it immediately shows how it generates so many files.

And my issue is always, how do I review so many files. Is the purpose just to go from 1 requirement prompt to so many spec files? How do I know the specs actually reflect the requirement?

I always feel I come away from projects like this even more confused.


Link clicked. Hijacks my scrolling. Link closed


I actually think I would like to use OpenClaw but everytime I go to the docs I give up.

It feels so intimidating. There's so much stuff going on. Maybe it is good for agents who can just motor through that volume, but me personally, I would want something simpler, maybe an `OpenClaw-lite`


OpenClaw is designed for people who are okay with "Don't worry about it" being an acceptable answer to "How does this work?"

Which is why it's so big with the PM/PO-turned-vibe-coder crowd.


But when the software is finished and works, who cares?

Not everything is security critical.

Like "Ohh you got a bunch of CRM addresses of people who go to a small time dog grooming company. You ready to sell that on the dark web?"


I'd assume anything empowered to read+send email from my personal accounts and read+modify my calendars to be security critical.


I guess you're not one of those "Don't worry about it" people


Hard to turn off when one's professional job/value is 'worry about things and find ways to efficiently minimize risk'


Use a vps and only give access to what you need.


Let's be honest, the number of people who are doing that << the amount of OpenClaw-preinstalled hardware that's being sold alone


Not my problem


When I tried OpenClaw it quickly turned into a lethargic mess which I really didn't see myself wanting to maintain long term. I looked at other alternatives which were similar enough to do what I was looking for (mostly scheduled tasks using data on my computer and some API data). I saw a mention of nanoclaw which I looked at but didn't try personally. I eventually settled on trying picoclaw[1] because it was extremely small, fast, and had enough features to do the job for me. It is under 10mb, fairly straight forward go codebase, and works with multiple model providers or locally hosted.

1 https://picoclaw.io/


I’ve concluded that if you already use a coding agent then openclaw adds very little (at least the multiple rounds I’ve tried it since mid Jan), it mainly brought the power of do-everything coding agents to your phone, available everywhere. But you exchange a low quality messaging interface for that ease of access.

Just to be clear, I’m not advocating that people that benefit from openclaw move to coding agents. This isn’t a “dropbox is just SFTP” comment, more that if you are already using coding agents, it feels like they do most of what I saw from openclaw at least a few months ago.

Am I missing something? Would love to be proven wrong.


It's not only that, all this complexity leads to stuff breaking with every update. For a tool that's supposed to just answer chat messages it's impressive how crappy and unstable this software feels. I have it running since the release and my primary usecase seems to be to use the CLI to fix whatever broke this week.

Would not recommend running this, it's the antithesis of the Linux philosophy.


Which features do you require?

I built an agent to work with local Ollama models, and it will keep working and iterate on a goal until finished, but it doesn't do sub-agents or or any of the "just burn a million tokens but finish this" pipelines that seem to be the trend now.

It's a simple D executable that basically consumes no resources in your local machine.


You may be interested in https://nanoclaw.dev/. It uses Claude by default but can run other harnesses after initial setup. Easier to grok IMHO.


There's a lot of similar projects out there that are closer to what you're looking for. For example, I used nanoclaw for a bit before switching to Hermes.


See if you like [Msty Go](https://msty.ai/go)


Yeah either the benchmark isn't very useful anymore or V4 Flash is a really, really good model.


In my use, DeepSeek v4 Flash (which replaced the quite excellent MiniMax M3) lags behind GLM 5.2 & Muse Spark 1.2 (let alone Kimi K3). Also, K3 is a much bigger multi-modal model, while Flash is text-only and likely optimised for coding tasks.


Yep, and the v4 flash final is about 2.5x slower than preview making it no longer a fast model, in fact slower than Luna and bigger models in many cases.

Spark is actually the interesting one imo. It's significantly better, also significantly faster. If you are ok with letting Meta soak up your data (which DS does too) it's also the same price.


GPT 5.6 Luna is an extremely cheap and still very capable model.

A chinese model being in the same ballpark of capability at half the price sounds believable to me.


It's significantly worse than Luna and quite a bit slower in some fairly involved tests I run.


That's fascinating, it's WAY better than luna ime. What sort of things are you testing it for?


I've been using this DeepSeek model the whole day today after building with 5.6 Luna extensively over the last week and I would disagree, at least for Rust + OpenGL.

DeepSeek just spend almost 2 hours trying to figure out why terrain textures were not working. It tried everything over and over again, it even had reference code for meshes on how to setup the rendering with materials, and it could just not do it.

I finally gave up and gave it to GPT-5.6 Luna instead, and figure out in a single prompt after 20 seconds, that the terrain mesh was being initialized with None in the material slot.

Other tasks it has managed to figure out at least, but it is significantly slower than GPT-5.6 Luna and it requires a lot more iterations.

(Both were set to high reasoning)


That's roughly my experience. Luna is extremely efficient and at higher levels of reasoning and longer running tasks more capable.

Reading the DS reasoning is wild, it's constantly going in circles. The most minor lack of clarity in your prompt and it will spend ages going back and forth on what you meant. It reasons 5x longer than the preview which makes it really slow now as well. We did a lot of work to nudge it to be decisive and improve our evaluation setup to there's more clarity, and it helped but only marginally.

Ours is a full-stack app one shot test so it includes backend, frontend, design, and QA/testing. It's graded by Opus xhigh and Sol xhigh and the grades are averaged.

DS4 preview would finish in 20 minutes flat on high reasoning and grades 6/10. Luna high gets 9/10 in about 30 minutes. DS4-final is crazy - at high thinking it's taking over an hour and getting ~8 but only had one successful run as I got tired of waiting so long after many early abort/retries trying to debug why thinking was so long. The lowest thinking still takes over 45 minutes, and with thinking off it actually is finally closer to preview in time but actually get's a much more varying result anywhere from incomplete to 6 it seems.

Costs per run DS4 is best but not actually by a lot as it's spending 10x the tokens with all the reasoning and mistakes. It's a very brute force model and I really preferred preview in many ways for how predictably fast it was.

Side note, Spark 1.2 is a nice model for this test, best in frontend design and fastest to get results together, though not nearly as efficient as Luna. Grok scores similarly to Spark but at like $50/run vs the contributor Spark costing $1.50.

Edit: was curious to see and seems DeepSWE agrees at least: https://www.together.ai/blog/deepseek-v4-flash-0731-vs-gpt-5...

Edit 2: btw it tests a team of agents working together in a special harness and stack. So 20 minutes is for 8 agents essentially. That said everything was built around DS as it was the cheapest to iterate against so even with that advantage the new one struggles.


Sorry, "Gemini 4 helped with editing"?


dang it i tried to check for a dupe but there were too many comments!


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

Search: