We celebrate each new model as a major step forward - but the inference layer has also improved dramatically in 2026.

Median agentic cache rate is up to 89% from the mid 60s to start the year.

Better cache = lower cost on the same model