When AI models achieve superhuman performance on specific benchmarks like coding challenges, it doesn't solve real-world problems. This is because we implicitly optimize for the benchmark itself, creating "peaky" performance rather than broad, generalizable intelligence.

Standard benchmarks are misleading for practical use. A model that benchmarks well can fail at agentic tasks. When selecting an open-source model, prioritize its documented ability to call tools and follow multi-step instructions, as this is crucial for building useful agents.

Traditional AI coding benchmarks are gamed or saturated. A new benchmark, DeepSWE, uses novel, complex tasks, revealing a massive performance gap where models like GPT-5.5 excel at 70%, while others trail by over 30 percentage points, contrary to other benchmarks that show them as close competitors.

A benchmark like SWE-Bench is valuable when models score 20%, but becomes meaningless noise once models achieve 80%+ scores. At that point, improvements reflect guessing arbitrary details (like function names) rather than genuine capability. This demonstrates that benchmarks have a natural lifecycle and must be retired once saturated to avoid misleading progress metrics.

Standard AI benchmarks are an engineering tool for measuring performance. A more scientific approach, borrowed from cognitive psychology, uses targeted experiments. By designing problems where specific patterns of success and failure are diagnostic, researchers can uncover the underlying mechanisms and principles of an AI system, yielding deeper insights than a simple score.

As benchmarks become standard, AI labs optimize models to excel at them, leading to score inflation without necessarily improving generalized intelligence. The solution isn't a single perfect test, but continuously creating new evals that measure capabilities relevant to real-world user needs.

Issues like 'saturation' and 'maxing' reveal a fundamental flaw: benchmarks test narrow, siloed abilities ('Task AGI'). They fail to measure an AI's capacity to combine skills to solve multi-step problems, which is the true bottleneck preventing real-world agentic performance and the next frontier of AI.

Early benchmark improvements focused on adding more languages and repositories. Now, the cutting edge involves creating more difficult evaluation splits through sophisticated curation techniques. Researchers must justify why their new benchmark is qualitatively harder, not just broader, than existing ones.

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.