No Wonder Anthropic and OpenAI are in a Rush to IPO
The rumor is that DeepSeek v5 will operate at Mythos’ intelligence level
The rumor is that DeepSeek v5 will operate at Mythos’ intelligence level. If true, I will bake it into Kilby, and Kilby into TEK2day Finance. Kilby needs frontier-level reasoning to navigate investment-related documents, spreadsheets and the investment scaffolding I’ve built on top of the underlying LLM (presently Anthropic models).
Anthropic’s Mythos model was trained on Nvidia’s Blackwell chips, which are already outdated. Nvidia’s Vera Rubin chip architecture is designed for agentic AI and long-context reasoning. Vera Rubin delivers up to 5x the inference performance compared to Blackwell for frontier LLMs. Rubin has almost triple the memory bandwidth of Blackwell (22TB vs 8TB), and houses 288 GB of memory per GPU, approximately 50% more per GPU than Blackwell.
I’m not sure when Rubin-level LLMs will come online, my guess is January.
China continues to distill frontier models (just as the frontier labs distill each others’ models). I am not sure how much longer the economics can hold for OpenAI and Anthropic if DeepSeek and others are going to come to market with similar or better models at a fraction of the cost (1/10th, 1/20th). No wonder Anthropic and OpenAI are in a rush to IPO.