The company representative said that they report all users that use Graphene OS, without any additional qualifiers. Presumably after they've already uploaded their personal details. That's the egregious part.
DeepSeek V4's KV cache is very efficient due to its heavily compressed and sparse attention architecture.
DeepSeek V3.2 which uses DSA only (sparse attention, but without compression from HCA and CSA) is a smaller model but uses 10x more memory at 1M context window compared to DS V4 Pro.
Also, I have to say, DeepSeek's API has a very good cache hit rate. With the same workload, I see ~80% KV cache hit rate with the DS API vs ~50% with the major western inference providers for open weight models.
I really hope Huawei ramps up Ascend production and DeepSeek open sources their optimized inference engine (they already open source a lot of their kernels -- kudos to them). This could shake things up.
Inference stack efficiency: Many of these providers take off the shelf sglang / vllm / trtllm and hope for the best. Meanwhile DeepSeek team is known for pushing the boundary of optimizations.
Now, sglang and vllm are great pieces of software, but take DeepSeek's Sparse Attention (DSA). Introduced 1.5 years ago (https://arxiv.org/abs/2512.02556), used by DeepSeek 3.2, GLM 5, DeepSeek V4. Only now is it slowly strating to get optimized in the major inference engines: (https://github.com/sgl-project/sglang/issues/19380https://github.com/sgl-project/sglang/pull/22851 etc.). Of course, DS V4 adds extra optimizations into the model architecture on top of DSA, and those will take more time to be taken full advantage of by the open source inference engines.
Privacy: Betting that people will pay extra for inference hosted outside China. This is especially true with DeepSeek, because DeepSeek is transparent about using API data for model improvements.
And few other things (scale (matters a lot for MoEs), reliability, soft enterprise lock in, etc.)
---
There is also, likely, tacit collusion at play here. Look at GLM 5 and GLM 5.1 prices. GLM 5 and 5.1 cost the same to run, but providers decided to charge much more for 5.1 because it is much better model, and because Z.AI raised their price as well.
I find it odd that sglang, vLLM, TRTLLM don't seem to want to publish benchmarks comparing each other. They used to, but now there seems to be some unspoken rule against it.
At least we get comparison against "other OSS engine" this time, but that could be HF's Transformers as well :)
> Again, your organization's Copilot interaction data is not included in model training under this new policy, but we are excited for you to enjoy the product improvements it will unlock.
Great work! There is maybe some bug. When you click on one of the 4 "opposing" countries (e.g. Czech Republic, Poland), it scrolls down and then shows that majority of the representatives from the country actually support it. Is that intended? Won't that make people from those countries "relax" even though they might have an impact by contacting their represenatives?