Irreversible: Anthropic Is Pushing Customers To Chinese Opensource LLMs
DeepSeek and others will take share from Anthropic and OpenAI
There’s a couple of reasons why I started to look at DeepSeek v4 this week - and I am far from alone in doing so - many Devs are moving to opensource alternative LLMs (the best open models are Chinese).
The reasons are:
1.) with the release of Fable 5, Anthropic decided it would store user prompts and model responses for 30 days even if the customer agreement said this would not be the case.
2.) more importantly, Anthropic unilaterally decided it would downgrade user access to its Fable 5 model without disclosing it if it deemed users were working on subject matter that was related to Anthropic’s core business. This happened to me when I was using Fable 5 to de-bug TEK2day Finance’s initial build. I was not working on anything AI-related, yet one minute I was inside Claude Code using Fable 5, the next minute when I checked which model I was using as something felt different, Claude Code showed I was operating Opus 4.8 without having ever selected that model.
What we use LLMs for in Kilby - reasoning through business problems, scoring transcripts for sentiment, building investment dossiers from disparate data sources, calling on tools and the like, I don’t know that we need a Mythos/Fable-level model for that. We likely will not rip out Anthropic from Kilby as I already finished the hard work of wiring Claude into Kilby’s infrastructure, but I will integrate DeepSeek v4 and subsequent DeepSeek models as an option by year end, and I very well may make DeepSeek the default LLM option.
Add to these reasons the fact that Anthropic clearly wants to be regulated as too big to fail, and you can understand why the developer community is pissed off at Anthropic. Anthropic knew the Government would pull its Fable/Mythos model. It will soon be restated. This is political theater. Next, there will be some agreement where the U.S. taxpayer is forced to invest in Anthropic. The Federal Government will subsidize Anthropic, which is now a quasi-utility like a power company or Fannie Mae. All of the user information from Anthropic’s models will be shared with the NSA/DoD just as cellphone data is mass collected.
The smart money is diversifying away from Anthropic. This is irreversible. Just as oil will never flow through the Strait again as it did in January, companies large and small, startups and giants, are diversifying away from Anthropic and OpenAI. Forever.