6 ms·
That is exciting! 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 aga
by bayesianbot 3mo ago
That is exciting!
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.
- hack1312 3mo agoWhat provider are you using?
- bayesianbot 3mo agoDeepSeek's own API
- greyb 3mo agoAny way to avoid China sales tax or is that just the cost of doing business?
- NortySpock 3mo agohttps://openrouter.ai/deepseek/deepseek-v4-pro#providers https://openrouter.ai/deepseek/deepseek-v4-pro#providers Look through the provider list for a company you are willing to do business with?
- verdverm 3mo agoFireworks.ai
- greyb 3mo agoI think I'd rather pay Chinese sales tax than work with Fireworks, but thank you for the suggestion.
- anigbrowl 3mo agoGood grief, the sales tax is only 6% on a service that's already extremely affordable.
- sudosysgen 3mo agoIf 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.
- yfontana 3mo agoHow come no other big model seems to be able to deliver the same type of extremely low cache cost though, if their techniques are public?
- petu 3mo agoDeepseek V4 paper is just ~three months old
- sudosysgen 3mo agoMany 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.
- jboss10 3mo agoI think the "architectural features" are part of the model, not the kv cache. So implementing it would be difficult and expensive.
- 3mo ago
- sourcecodeplz 3mo agoit is ridiculous really. it is so cheap, that i can just run it basically 24/7.