Overview

• OpenAI restored access to retired models—including GPT-4o, GPT-4.1 and o3—for paid subscribers following intense user backlash over GPT-5’s colder, shorter responses.
• The company identified and patched a router glitch that had misrouted queries to weaker engines and degraded GPT-5’s performance.
• Paid subscribers can now select per-query modes—Auto, Fast and Thinking—with a weekly message cap applied to GPT-5’s Thinking mode.
• Enterprise clients and developers have doubled their GPT-5 API usage for coding and reasoning tasks since launch, reflecting strong business demand.
• Some analysts and users believe GPT-5’s unified router architecture was driven by compute-cost savings rather than solely by capability improvements.