Directly at DeepSeek? It was my understanding (but I didn't check) that some other AI operators were providing (some of?) DeepSeek's model for cheaper prices.
Still, that's interesting. What do you get for that price? Only coding, or also e.g. image generation?
DeepSeek API gave 6x to 8x better caching rate for inputs over OpenRouter (even chosing DeepSeek as provider). And some of the cheaper providers are using FP4 quantizations.
After complaints the cached read is not listed anymore in that page, you have to click one by one. All providers for DeepSeek V4 Flash charge ~$0.02 while DeepSeek provider is $0.0028. For coding this is huge as caching often gets in the range of 90 to 99%. But OpenRouter messes your caching so don't use it. And it seems to be a VC-backed closed middle-man company, not open source or open anything.
Footprint's comment is correct. I go directly to Deepseek's platform API which they linked. There's no image generation but you get access to Deepseek V4 Flash and Deepseek V4 Pro, both of which are very capable for general text based tasks and programming. Flash is insanely cheap for how good it is ($0.14 per 1M input tokens vs $15 with Claude 4.7). V4 Pro I would put somewhere in the range of 80 to 90% as good as Opus 4.6 (based just on anecdotal usage - I use Opus 4.6 heavily at work as my company pays for it) while again being significantly cheaper. According to a benchmark[1] I read, processing 1million tokens would cost you $250 for Opus 4.7, $300 on GPT5.5... and just $35 on V4 Pro.
I just use it for my side-project coding and brainstorming tasks. At work I use AWS's Kiro CLI + Opus 4.6. At home I use Opencode + V4 Flash for the majority of "general" usage. I swap to V4 Pro for complex tasks if I feel like V4 Flash is struggling.
One other thing I highly like about the platform.deepseek API usage is it's a metered setup - not subscription based. Which means you only pay for what you use (the money that you put in doesn't expire) and can't spend more than you've deposited. This works well for me for my non-work coding because it generally happens in bursts. I may not code for a whole month (and therefore if I had a subscription it would have been wasted) and then spend a whole weekend coding nonstop.
It's entirely possible that there are middle-man providers that give a discount on Deepseek's own pricing, but I'm quite happy with the amount I'm paying so I haven't really looked into it.
I’ve been doing this too, it’s a cheat code! 1/100th of the price of Claude/openai prices for 95% of the quality. Site is platform.deepseek.com for that. No image generation, just text, but if you use it right it works great
I'm assuming that you need to feed the human being (i.e. you) regardless of whether you use that human being for writing code or not. So, by this metric, there is simply no breaking even point. The cost of human + AI is always going to be higher than the cost of human.
Anecdotally, a few weeks into a Rust agent-first project, we're still trying to get the agent to maintain a minimum of coding discipline (e.g. don't use sync Mutex in tokio code). So far, the agent seems more interested in deactivating the linters than in complying.
Security? At this stage, I'm a bit afraid that it's a joke more than anything else.
That should be solvable by denying permission to edit the lint files with a message saying lint files cannot be edited and not to use workarounds (sed, scripting etc)
You could also use hooks to block running of scripts for some number of turns after an attempt to cheat.
> Why on earth would AI labs be bragging about how little the product they sell actually costs them to make? You don't want to do anything that reduces it's perceived value to the user, that might make them less willing to pay for it.
Wouldn't they be bragging about it to investors? It feels like something that would matter a lot to them, and at least OpenAI kinda feels desperate to find them.
There's also the small question about whether a drop in inference cost would actually change anything about profitability, when training seems to get exponentially more expensive.
It seems that this era is a marketing experiment for Mythos.
We're running forward without any idea of how we can get agents to write code that is even remotely safe or secure. It _will_ blow up with increasingly large blast radiuses.
Go is pretty good at performance, but pretty bad at expressing domain-specific logics. Python is the opposite, but once you have isolated the parts that need to be optimized, it's quite easy to rewrite them in a native language (in particular, the Rust-Python bindings are really good, although in this project, it's C++).
Pepper was a pun on Native Client (since NaCl = salt). Pepper Plugin API (PPAPI) was Google's more secure version of NPAPI (Netscape Plugin API). Flash Player was essentially the only thing using NPAPI/PPAPI by the end of its life.
The most common plugins were Flash, Silverlight, Adobe Reader, and the Java applet plugin, and I think all of those were in mildly common use when plugins were on their last legs.
When I was a manager in a start-up, ages ago, I argued the CEO against handing a (small) one-off bonus to one of my team members, and rather went shopping for a nice gift with the same sum. One of them was purely a transaction, the other one was a gift.
Still, that's interesting. What do you get for that price? Only coding, or also e.g. image generation?
reply