Recent Developments in Model Usage
On March 27, 2026, Anthropic’s Thariq Shihipar revealed significant changes to the usage policies for the Claude model. The company will now throttle user sessions during peak hours, specifically from 5am to 11am PT. This adjustment affects all subscription tiers, including the $100 per month Max plan, leading to accelerated consumption of the 5-hour session limits. According to Shihipar, about 7% of users, particularly those on Pro tiers, will likely encounter these new restrictions.
OpenAI seized the moment to capitalize on Anthropic’s decision. On the same day, Thibault Sottiaux announced that OpenAI would remove caps on Codex usage across all plans. This strategic move allows users to experiment freely with newly launched plugins, contrasting sharply with Anthropic’s restrictive policies. The timing of these announcements suggests a competitive maneuver to capture users dissatisfied with the new limitations imposed by Anthropic.
Implications of Anthropic’s New Policies
These session limits raise operational concerns for businesses reliant on Claude for their workflows. Users accustomed to seamless access during peak hours now face interruptions, forcing them to adjust their usage patterns. Recommendations to run intensive tasks during off-peak hours reflect a fundamental misunderstanding of how users engage with AI tools; most do not plan their sessions around arbitrary time slots.
Moreover, the opacity surrounding usage metrics presents a risk for businesses. Anthropic’s model, while boasting weekly limits, fails to clarify how quickly users will hit these caps. This lack of transparency could lead to unexpected disruptions, particularly for enterprises that depend on consistent access to AI models for production-level tasks.
OpenAI’s Strategic Positioning
OpenAI’s removal of usage caps on Codex signals a calculated effort to attract users disillusioned by Anthropic’s changes. By enabling unlimited experimentation, OpenAI positions Codex as a more appealing option for developers and businesses alike. This approach allows OpenAI to gather user data and assess demand before imposing any future constraints or pricing adjustments.
The competitive dynamics in the AI model landscape are intensifying. With OpenAI offering GPT-5.4 at competitive rates—$2.50 per 1,000 tokens for input and $15 for output—Anthropic’s pricing of $5 and $25 for Claude Opus 4.6 appears less attractive. As demand surges, this scenario sets the stage for a cycle of price adjustments and usage limits, where companies jockey for market share.
Conclusion: A Shift in the Competition
The situation underscores the evolving nature of AI model pricing and usage policies. As businesses navigate these changes, understanding the mechanics behind pricing and access will be crucial for making informed decisions. The rapid response from OpenAI indicates a recognition of the competitive landscape and a willingness to adapt to user needs, setting a precedent for future interactions in the market.
