The only AI benchmark that matches my experience

Artificial Analysis is so far the only benchmark that has matched my real-world experience with frontier AI models.

I was a long-time Claude user (since Sonnet 3.5), well before the Anthropic hype, but canceled my $20 subscription about 6–7 months ago due to rate limits. For me, using my own cognitive "stack" to divide large problems into smaller bits and then sending them to a quite-competent assistant is much more fun than complete cognitive surrender to AI gods that one-shot complex tasks and then exploding my quota after failing tests.

Why not use something dirt-cheap and 80% as capable while keeping your cognitive functions alive? That's how I interpret the main figure on this webpage.