OpenAI Wants Perplexity to Become the Shared Yardstick for AI Progress

Perplexity and Intelligence

OpenAI is calling for perplexity—a measure of how well a model predicts the next token—to become an industry-wide benchmark for tracking AI progress and setting safety limits. The company is publishing its own internal perplexity numbers, arguing that shared transparency outweighs competitive risk. The post traces how capabilities emerged as perplexity fell from GPT-3.5 (~0.99) to GPT-4 (~0.91) to below 0.8, and predicts further breakthroughs in the ~0.68–0.7 range.

For many of us that have previously worked in mathematics and adjacent fields, seeing the resolutions to questions we’ve long thought about coming out of the model one after another in a single Slack thread felt deeply surreal.

More from this day

2026-11-22