They have pretty much lost already. The US will probably try to fight in some way, but they have a very little moat. Even if they actually “stole”, it’s not as if that Anthropic had the moral high ground here.

Some chinese models like GLM 5.2, Kimi K2.7 Mimo V2.5, Deepseek V4 Pro and Minimax 3 cost close to nothing and have wild usage limits if you subscribe.

You can also run these Chinese models on European infrastructure through Cortecs.ai or similar. And you have actually have privacy! Privacy + Cheap vs Expensive and slightly better.

Opus 4.6 is almost 5€ in and 24€ out. Sonnet 4.6 is 3€ in and 15€ out. Minimax 3 is 0,3€ in and 1,7€ out.

All frontier models are so good that the differenciator going forward is going to be price. The models keep improving, but there is a clear trend of diminishing returns.

I tried showing my friends how Gemini is pretty balanced and not as sycophantic as the rest. After 5 queries they seemed quite surprised of how balanced and nok sycophantic it was.

At the sixth request is said: “based on your instructions, I should not simply agree with you.”

Turns out I had a pretty substancial set of instructions to make it useful and had forgotten about it 😂 Explains why I have preferred Gemini for a long time.

For good reason, there is a lot of interest in implementing actual bias.

Israel/Palestine/Gaza is examples of this.

Do you have any examples of bias? Or wrong information?

How is it biased? Biased towards truth? If you have proof of it lying or misrepresenting something, then you just have to prove it and I am sure they or you will be able to change it.