I think gpt 5.6 family also dropped pricing but didn't give any more usage for the subscriptions. Maybe it's a way to silently lower the value given to subscriptions while keeping API pricing competitive
1M cached tokens on deepseek is $0.006, the big labs can't sell anywhere close to this, they have funders expecting returns and huge overhead.
btw I've had a ton of fun with the new deepseek today, I was waiting for my OpenAI 5h limit reset and decided to give it some problems for fun, got pretty great results. Tried some harder problems and still got great results. I don't expect it to be Sol class or anything but I really didn't expect it to be anywhere near this good so we'll see where it ends up. And it's really fun throwing crazy amount of tokens at the wall for ~free instead of watching the subscription limits tick closer while your agents churn away.
> 1M cached tokens on deepseek is $0.006, the big labs can't sell anywhere close to this, they have funders expecting returns and huge overhead.
Losing most of your customers tends to sharpen the mind a bit. They could eg stop pushing out the absolute frontier for a while and focus on making what they have run cheaper. Or they go and do more lobbying against China. Or a million other little things that take more than 30 seconds to come up with when writing a HN comment, but less than a week for someone who's smart and paid to do this for a living.
I did a lot of that kind of work with the older version of DeepSeek before they upped the prices.
For example, it was quite good to get a decent Sashiko review. Sashiko is a Linux kernel review agent with interchangeable LLM driver. It's very good, but it eats tokens like crazy.
This clearly is Codex-specific, not as much a feature of the model (though obviously they probably have trained it to be great at working with their own tools). Sounds somewhat similar to pi-observational-memory I'm using with Pi.
Maybe you could try go2rtc[0]? It can ingest multiple stream formats and output them in a different one.
Also has some other nice features, like I use it to ingest one RTSP stream and share that to multiple applications, and to push audio out to my Tapo camera
Are they related? The twitch stream definitely didn't have nudes while I was watching (while the other link seems to be ~constant nudity). Also the twitch description was something like "With new/fixed content moderation" or something similar (so might be right that they've had moderation problems before)
IIRC GPT 3 was priced at per 1k tokens, had to check, the biggest GPT 3 model from OpenAI was $0.06/1k, so $60 / 1M. gpt-3.5-turbo was the first model after ChatGPT and that was $2 / 1M. And no caching. So not really in the same ballpark
I think for me customizing my setup isn't time away from "real tasks" - I do the customization when I don't feel like doing the work or I'm doing some background thinking on how to proceed in the real task. It's an interesting way to keep my mind warmed up for programming through times when I'm not in the mood, and I'd guess 80-90% of the customization is waste of time, but the ones that I end up sticking with can be really useful and save time/effort dozens of times per day
I don't understand how DeepSeek can be so cheap with their cache pricing - ~0.003 usd / 1Mtok. 100x less than Kimi K3, or similar numbers against pretty much any other decently sized model to my knowledge. I've been using it whenever possible as even longer agent sessions cost few cents.
If you read DeepSeek's papers, you'll find a litany of architectural features that allow for a greatly reduced cache hit price by shrinking the size of the KV-cache.
Many of these techniques haven't been published very long ago - it often takes a good 6-8 months for techniques to percolate. But also, they come at a complexity cost and, seemingly, also at a stability cost.
Also potentially a performance (in terms of output quality) cost. DeepSeek is cheap on a per token basis but lags behind in the benchmarks, perhaps it was a calculated tradeoff.
reply