Deepseek V4 is an open-source language model with a 1 million-token context window costing $1.74 per million input tokens, compared to GPT-4.5 and Claude Opus at much higher rates. Companies can run it on their own servers for security and privacy, potentially disrupting the AI market by making state-of-the-art capabilities affordable.
Deepseek V4 is nearly state-of-the-art. It's open sourced and it's got a 1 million token context length. The biggest differentiator and the reason all of the big model companies sort of freak out when models like this come out, it sort of has a little bit of a shock on the stock market is because of this pricing. This is basically almost as good as the best models out there and it cost $1.74 per million tokens input and $3.48 per million tokens output. If we compare that to GPT-5.5, GPT-5.5 is $5 per 1 million tokens input and $30 per 1 million token output. And even GPT-5.4, $2.50 per 1 million token input, $15 per 1 million token output. And this model is roughly as good as GPT-5.4. If we compare it to Claude Opus 4.7, $5 per million token input, $25 per million token output. And even 4.6 here is $5 per million token input and $25 per output. And again, it's kind of on a similar level to Opus 4.6. So with that in mind, companies are able to get nearly the same capability at a significant discount. And because these models are open, they could theoretically put them on their own supercomputers in their offices and run them locally and have all of the security and safety and privacy benefits of having these local models.
AI News: Dots, GPT-6.1 Sol, Sonnet 5.5, Gemini 4, and everything you need to know
