The article analyzes the latest coding models, Opus 4.6 and Codex 5.3, highlighting their usability and performance differences. Codex 5.3 shows significant improvements over its predecessors, but still lags behind Claude in user-friendliness and overall experience. The discussion also touches on the shifting importance of benchmarks in evaluating AI models.

This article analyzes how benchmark scores for AI models often reflect a single dimension of "general capability." It discusses the implications of this finding, particularly the contrasting ideas of whether model performance is based on a deep underlying ability or if it is contingent on specific skills. The author also introduces the concept of "Claudiness," which reveals limitations in certain model capabilities.

The article reviews GPT-5.2, highlighting that while it has notable improvements in instruction-following and complex task handling, its performance is slower than expected. The author compares it to other models like Claude Opus 4.5 and Gemini 3, noting that it may not be the best choice for all use cases, especially in coding or when a more engaging personality is desired.

The article presents benchmarks for text-to-image (T2I) models, evaluating their performance across various parameters and datasets. It aims to provide insights into the advancements in T2I technology and the implications for future applications in creative fields.

The performance of the gpt-oss-120b model on private benchmarks is notably worse than its public benchmark scores, dropping significantly in rankings, which raises concerns about its reliability and potential overfitting. The analysis suggests a need for more independent testing to accurately assess the model's capabilities and calls for improved benchmarking methodologies to measure LLM performance comprehensively.