I've been using DeepSeek exclusively since I've been playing with LLMs, largely due to the price. I joined a company that is an Anthropic shop and I am genuine

I've been using DeepSeek exclusively since I've been playing with LLMs, largely due to the price. I joined a company that is an Anthropic shop and I am genuinely shocked. Sonnet 5 is a little better on long horizon tasks and headless unsupervised agent workflows - but for in-IDE workflows, it's virtually unusable. I am so used to flipping around my codebase at warp speed with DeepSeek flash. It's so fast and accurate I don't have time for parallel agents. It's a really rewarding workflow. Moving to Sonnet, you ask is something simple like "split this into a seperate file" "implement this method" "this is my schema, implement a repository for it". It'll spend 30 minutes thinking and charge like $12. And no token caching, what are you even doing Anthropic? It's unusable. DeepSeek are in a league of their own