AI models show impressive performance on evaluation benchmarks but underwhelm in real-world applications. This gap exists because researchers, focused on evals, create reinforcement learning (RL) environments that mirror test tasks. This leads to narrow intelligence that doesn't generalize, a form of human-driven reward hacking.

OpenAI's evals team is looking beyond current benchmarks that test self-contained, hour-long tasks. They are calling for new evaluations that measure performance on problems that would take top engineers weeks or months to solve, such as creating entire products end-to-end. This signals a major increase in the complexity and ambition expected from future AI benchmarks.

AI struggles with long-horizon tasks not just due to technical limits, but because we lack good ways to measure performance. Once effective evaluations (evals) for these capabilities exist, researchers can rapidly optimize models against them, accelerating progress significantly.

Standard benchmarks are too rigid. The future of model evaluation needs more open-ended, multi-agent scenarios like the "AI Village" project. Giving agents broad goals like "organize an event" reveals more about their "derpy" failure modes and real-world capabilities than constrained, benchmark-style tasks can capture.

Just as standardized tests fail to capture a student's full potential, AI benchmarks often don't reflect real-world performance. The true value comes from the 'last mile' ingenuity of productization and workflow integration, not just raw model scores, which can be misleading.

Traditional AI benchmarks are seen as increasingly incremental and less interesting. The new frontier for evaluating a model's true capability lies in applied, complex tasks that mimic real-world interaction, such as building in Minecraft (MC Bench) or managing a simulated business (VendingBench), which are more revealing of raw intelligence.

Alex Karp argues that an AI's high score on a single benchmark is irrelevant for enterprise adoption. Real institutions require passing thousands of consecutive, differentiated tests. An AI model that is brilliant at one task but fails at the 50th in a complex sequence is effectively useless.

OpenAI identifies agent evaluation as a key challenge. While they can currently grade an entire task's trace, the real difficulty lies in evaluating and optimizing the individual steps within a long, complex agentic workflow. This is a work-in-progress area critical for building reliable, production-grade agents.

Obsessing over linear model benchmarks is becoming obsolete, akin to comparing dial-up speeds. The real value and locus of competition is moving to the "agentic layer." Future performance will be measured by the ability to orchestrate tools, memory, and sub-agents to create complex outcomes, not just generate high-quality token responses.