GPT-5.2 Can't Identify a Serial Killer & Was The Year of Agents A Lie? EP99.28-5.2

OpenAI’s GPT‑5.2 hit the market with a 400 k token context window and a modest price bump to $1.75 per million input tokens. While the larger window and vision tool upgrades sound impressive, most users report that the core model feels unchanged, delivering overly verbose answers that add little value. The release feels like a tuning exercise rather than a breakthrough, and the higher cost raises questions about ROI for businesses seeking efficient, concise outputs.

The biggest pain point emerges in tool‑calling and agentic workflows. GPT‑5.2 often fails to chain multiple tool calls, stumbling on parallel requests and refusing to self‑correct when a call goes awry. In contrast, Anthropic’s Claude models operate with an internal “clock,” iterating over tasks and allowing users to intervene between rounds. This iterative design yields more reliable agentic behavior, especially for complex, multi‑step processes. OpenAI’s newer ethical guardrails also inject unnecessary disclaimer language, further slowing down interactions and limiting practical utility.

Competitive pressure is reshaping the landscape. Grok 4.1 offers a fraction of the cost—around $0.20 per million tokens—and delivers faster responses, making it attractive for time‑critical tool‑calling scenarios. Gemini’s integration into Google services is gaining user trust, while DeepSeek’s emergence broke the perception that ChatGPT is the sole AI option. These dynamics have inflicted brand damage on OpenAI and delayed the anticipated “year of agents.” As enterprises experiment with delegation versus iterative collaboration, the market will likely reward models that balance speed, cost, and reliable agentic capabilities.