OpenAI’s recent GPT-5 release has sparked fervent debate within the AI community, with initial expert assessments suggesting a powerful leap in utility and accessibility, rather than an immediate arrival of Artificial General Intelligence. On a recent bonus episode of the Mixture of Experts podcast, host Bryan Casey convened with Distinguished Engineers Mihai Criveti of Agentic AI and Chris Hay, CTO of Customer Transformation, to dissect the implications of OpenAI’s latest models.
Despite considerable anticipation, GPT-5 has not universally dethroned existing models for specialized tasks like coding. Chris Hay candidly stated, "Sadly, I had high hopes, but no," when asked if GPT-5 would replace Claude as his primary coding tool. Mihai Criveti echoed this sentiment, revealing, "I found myself using Claude code and Opus 4.1 again. I think it's an amazing model, but right now it's not replacing it yet for me."
However, GPT-5 brings significant improvements in reliability, particularly "reductions in hallucinations," making the models more trustworthy for diverse applications. Beyond raw intelligence benchmarks, Bryan Casey underscored the strategic introduction of a model router and highly competitive pricing across the GPT-5 series. The pricing strategy, which he described as approaching a "too cheap to meter" reality, marks a pivotal shift towards democratizing advanced AI.
A key differentiator for GPT-5 lies in its enhanced agentic capabilities. Mihai Criveti emphasized, "I think what stuck with me this release is how good this model seems to be at tool invocation and calling for AI agents," attributing this to precise fine-tuning for structured outputs and function calling. Chris Hay highlighted the "Nano model" on the API, which "outperforms most of the large models in the market" for agentic tasks, suggesting a future where smaller, highly specialized models become increasingly powerful and efficient. This accessibility and specialized performance of smaller models could accelerate AI experimentation and integration across various industries. The overall landscape suggests a future of diverse, highly capable models, each excelling in distinct domains, rather than a single, all-encompassing AGI.
