The launch of OpenAI's GPT-5 has ignited a remarkably polarized reaction across the artificial intelligence landscape, as meticulously chronicled by AI commentator Matthew Berman. His comprehensive review of industry feedback, spanning direct statements from OpenAI's CEO to independent benchmark analyses and developer sentiments, paints a picture of conflicting perspectives on the model's true impact.
Matthew Berman provided a sharp analysis of the industry's varied reactions to OpenAI's GPT-5 release. He dissected a multitude of social media posts and independent evaluations, revealing a landscape of both fervent praise and critical skepticism surrounding the new model.
A core insight emerging from this reception is the nuanced understanding of "progress" in AI. Sam Altman, OpenAI's CEO, candidly admitted, "We for sure underestimated how much some of the things that people like in GPT-4o matter to them, even if GPT-5 performs better in most ways." This highlights a significant realization: raw performance metrics, while crucial, do not fully capture user satisfaction or established workflows. The familiarity and perceived "warmth" of previous models unexpectedly factored into user adoption, prompting OpenAI to consider future customization options for different user personalities.
Independent benchmarks, however, tell a compelling story of technical advancement. Artificial Analysis, for instance, declared, "GPT-5 sets a new standard with a score of 68 on our Artificial Analysis Intelligence Index," with its "High" reasoning effort reaching a new intelligence frontier. This was echoed by LMarena, which positioned GPT-5 as the top model across various categories including text, web development, and coding. Yet, some evaluations, like those from Stagehanddev, presented a counter-narrative, suggesting GPT-5 performs "worse than Opus 4.1 in Stagehand evals in both speed and accuracy" for certain tasks.
This divergence underscores a critical shift in how AI professionals evaluate models: the focus is moving beyond mere benchmark scores to practical utility. As Theo, a prominent developer, asserted, "I don't care about 'intelligence' benchmarks now. I'm post-eval. GPT-5 does what you tell it to do. No other model behaves this well." This sentiment emphasizes the importance of a model's instruction following, its ability to integrate with tools, and its overall "vibe" in real-world applications. The enthusiasm for GPT-5's ability to "one-shot" a Minecraft clone or refactor an entire codebase in "one call" (even if the refactoring didn't always work, as one developer humorously noted) speaks volumes about its perceived agency and practical capabilities.
