Looks like it's the same price as Argon on openrouter, which is good to see. Probably was announced already somewhere that I missed.
It's interesting that Opencode Go got Argon, I'm hoping they'll get Carbon as well. They claim that Argon is ZDR on their platform. That's surprising to me because openrouter doesn't have ZDR endpoints for Google's models any more, due to a change in policy.
Deepseek is good enough for me that I use it for everything cost sensitive. I tend to switch to Google's models for world knowledge, planning, and complicated bugs that Deepseek gets stuck on. Argon is also a great writer in my experience, hopefully Carbon is the same. Seems like RL for code generation ruins most model's communication abilities. V4.1 Flash 1017 is a significant step back from the previous Deepseek model in that respect, despite being a noticeably better coder.
I don't think I've seen opencode change the system prompt of a conversation, nor pi (though I have less experience there). Is there a specific scenario you're referring to where this could happen?
That's an old, closed issue. I think this has worked since the release of OpenCode v2, it looks like they added the ability to send update messages for the environment here: https://github.com/anomalyco/opencode/pull/30789
I haven't noticed any cache busting happening at midnight, but I'll pay more attention in the future just in case.
Of course the lack of any care for submitted issues or follow up from the devs is a long term known issue with Opencode, but it's not what we're talking about here.
Could you try taking a musical track and doing a low-pass filter to remove anything above 100 hz? If it doesn't sound about the same to you, then that is proof that you are hearing and processing some sound above that 70 hz limit. The inability to distinguish speech sounds more like trouble with higher frequencies (which help discriminate between different syllables) than with frequencies in the 100-400 hz range typical of human speaking f0s - because the latter are also vital to what tends to make music sound full and rich.
The AA benchmarks seem to be mostly "complete this small intelligence-proving task from scratch". That consistently puts models with high intelligence to output tokens on the frontier, especially the OpenAI models, on the frontier. But a lot of software engineering work is many-turn work on established code bases. There's a lot of reading and a relatively large context unless it's carefully controlled. Writing only a few output tokens doesn't really cut costs because most of the spend is read tokens that get recycled through dozens of turns. That's basically the ideal case for DeepSeek because of how cheap cached tokens are.
The thing I'd be worried about when doing this is spending $1000+ on a TV, then realizing it has some sort of pre-downloaded ads that it shows on the main screen ("sign up for your free six months of HBO Max!") or otherwise has a nag prompt that appears to tell you to connect to WiFi. I think that would be basically intolerable.
Been a few years since I had to get a new one, but I booted mine up, switched to the HDMI input for my AppleTV, and never looked back. Just boots straight into the AppleTV input when you use the AppleTV remote to wake it up.
We have to use the actual TV remote to debug some audio issues or disable the motion blur stuff again every once in awhile, but otherwise it might as well just be a display for the AppleTV. I’ve only seen the actual home screen of the TV a handful of times.
I guess it’s probably still spying on us somehow, and it seems like different models make using the AppleTV remote and booting right into the AppleTV harder than others, so your mileage may vary.
You have to be careful with this path. Some manufacturers embed LED's into the bezel and blink them at you (part of the IR sensor for the remote, to prevent you from taping over with electrical tape) if they do not have network.
I think the places the average person starts out with AI do look like this. Most people are not paying a $20 a month Anthropic subscription just to try it out. They're using free chat apps, reading the garbage llm summary Google puts at the top of every search page, dealing with the low quality tools foisted on them in their email app or word processor.
Have all these gotten a lot better in the last year? Sure! But there's still a several-orders-of-magnitude difference in quality between whatever garbage tier of GPT-6 Luna you can use for free on `duck.ai` (just for one example) and Opus 5.5. Or even GLM 5.3 Flash. It's a way bigger difference than the improvement in ~free models has made up for in the last year, even though those improvements have been substantial.
I think a certain sort of person, if they do try to use something better along with a coding harness, as also immediately inclined to read along with the reasoning traces - and understandably gets freaked out by the nonsense it contains. That's a mistake, I think; not just because models behaving "reasonably" doesn't depend on the rationality of their reasoning traces, but also because I think if you transcribed my brain activity into a serial stream of text, I'm not convinced it would be more coherent.
Maybe I'm misunderstanding you, but a better attack on a reduced-round AES has been achieved by LLMs. Anthropic did it and published it already, a chosen plaintext attack that improves on state of the art against reduced-round by "200-800x". https://anthropic.com/document/aes_mobius_bridge.pdf
The only reply to this criticizes the hedging, but I think it's actually true and it does make it harder to tell if you're vulnerable to this. And on top of that, quite a few systems have brotli installed by necessity. On Debian 12, for example, I have the following dependency chain on a VPS I run:
Argon was announced 30 September, publicly available 13 October. Carbon announced 28 October, publicly available 10 November.
reply