OpenAI's '0'Ultra' Reasoning Level Sparks Controversy Over Hidden Costs and Sub-Agent Implementation

OpenAI’s recent GPT 5.6 release, featuring new models like Sol and Terra, has ignited significant debate, particularly regarding its new “Ultra” option. While positioned as a superior reasoning level alongside “Max,” critics contend that Ultra is fundamentally misrepresented, functioning instead as a toggle for an advanced but flawed sub-agent system. This implementation, distinct from a true reasoning level, has been criticized for causing considerable confusion among users, leading to inefficient resource utilization and unexpectedly high token consumption.

The crux of the issue lies in Codeex’s V2 sub-agent implementation, which Ultra leverages. This V2 system defaults to sharing the full historical context with all spawned sub-agents, significantly increasing token burn and risking context pollution. Furthermore, V2 allows for infinitely nested sub-agents without a built-in depth limit, creating a high potential for uncontrolled execution and exorbitant costs, with users reporting the exhaustion of 5-hour rate limits in minutes. This behavior stands in contrast to Anthropic’s “Ultra Code,” which utilizes a “workflow” approach where sub-agents are programmatically defined with distinct phases and termination points, offering greater control and predictability. Critics suggest OpenAI has adopted the less effective user experience of presenting Ultra as a reasoning level without incorporating the robust implementation of controlled workflows.

Despite some real-time UI adjustments by OpenAI in response to feedback, the underlying sub-agent architecture remains a concern. The temporary removal of OpenAI’s 5-hour rate limit further compounds the risk, potentially allowing Ultra to deplete weekly quotas much faster without the previous soft warning. The community advocates for a redesign where Ultra is treated as a distinct “skill” or “switch” rather than a reasoning level, coupled with more robust, workflow-driven sub-agent management that allows for setting individual reasoning levels and clear termination conditions. Users are advised to exercise caution with Ultra or consider alternative programmatic sub-agent strategies.