INTELLIGENCE YIELD

Summary from the TLDR AI newsletter

Everyone focuses on how hard a task can models solve. Benchmark data also captures how much work models need to get there. Opus 4.6 solves harder tasks, more reliably, with considerably less compute. The post contains charts comparing models from Anthropic, OpenAI, and Google.