OpenAI tests ChatGPT Agent upgrades powered by new Alpha models

Some OpenAI users have recently spotted a new “alpha models” section in the ChatGPT model selector, briefly showing experimental agents labeled with terms like “Agent with truncation” and “Agent with prompt expansion.” These models appeared for a limited window and both activated the agent mode, meaning ChatGPT would attempt to use a browser or additional tools automatically to complete user tasks. The distinction in naming, truncation vs. prompt expansion, suggests that OpenAI may be experimenting with different system prompt setups or even underlying model architectures, potentially as a staging ground for GPT-5-powered agents, given that previous agent releases used earlier model versions.

The new agent variants would mainly appeal to early adopters, researchers, and power users interested in tool-augmented AI workflows and those tracking the evolution of autonomous AI agents. The exact improvements or limitations are still unclear, as the release seemed accidental and was quickly rolled back, making comparison with existing agent modes challenging.

The “alpha models” were surfaced in the core ChatGPT app, likely as a backend toggle rather than a UI-controlled feature, and quickly disappeared. This aligns with recent comments from OpenAI CEO Sam Altman, who hinted at major product announcements and feature releases in the near future. Such tests typically signal final-stage development before a broader rollout. If these agent models move beyond testing, they could change how tasks like browsing or research are handled, perhaps introducing new strategies for managing context limits (truncation) or prompts (expansion).

The discovery was first made by users actively monitoring the model drop-down and experimenting with tool usage. While no official statement has clarified the exact intent, it’s evident OpenAI continues to iterate quickly, likely preparing for a more robust and flexible agent launch as competitive pressure in the AI toolchain space grows. This would fit with OpenAI’s ongoing push for agents that blend reasoning, browsing, and other skills natively inside ChatGPT, closing gaps with products like Copilot and Perplexity’s Comet.