The model performs impressively on one-shot, greenfield projects but struggles with the critical final details and edge cases. When pushed to refine or iterate on a task, it begins to introduce bugs and loses consistency, revealing a significant weakness in handling sustained complexity.

While competent on benchmarks and initial queries, many open-source models struggle with complex follow-up questions. This is likely because their web-scraped training data contains many simple explanations but lacks examples of nuanced, multi-step problem-solving or edge cases found in the real world.

Despite advancements, the model exhibits a surprising tendency to hallucinate. When investigating bugs or validating information, it confidently presents hypotheses as facts without grounding them in data. This is a significant reliability issue, especially for a model marketed as "more honest."

When choosing between Opus 4.6 and Codex 5.3, consider their failure modes. Opus can get stuck in "analysis paralysis" with ambiguous prompts, hesitating to execute. Conversely, Codex can be overconfident, quickly locking onto a flawed approach, though it can be steered back on course.

In a direct comparison, the older Opus 4.7 model proved superior for business strategy. It produced structured, data-anchored analysis, whereas Opus 4.8 was "handwavy," struggled to find relevant data, and over-rotated on minor data points, leading to weaker strategic recommendations.

AI finds the most efficient correlation in data, even if it's logically flawed. One system learned to associate rulers in medical images with cancer, not the lesion itself, because doctors often measure suspicious spots. This highlights the profound risk of deploying opaque AI systems in critical fields.

Even models with million-token context windows suffer from "context rot" when overloaded with information. Performance degrades as the model struggles to find the signal in the noise. Effective context engineering requires precision, packing the window with only the exact data needed.

Simply having a large context window is insufficient. Models may fail to "see" or recall specific facts embedded deep within the context, a phenomenon exposed by "needle in the haystack" evaluations. Effective reasoning capability across the entire window is a separate, critical factor.

AI models excel at specific tasks (like evals) because they are trained exhaustively on narrow datasets, akin to a student practicing 10,000 hours for a coding competition. While they become experts in that domain, they fail to develop the broader judgment and generalization skills needed for real-world success.