OpenAIDevs Announces /fast Mode: GPT-5.4 in Codex Runs 1.5x Faster

Codex speed update on X

OpenAIDevs posted on March 5, 2026 that Codex now has a /fast mode. According to the announcement, GPT-5.4 can run about 1.5x faster in this mode while keeping the same intelligence and reasoning behavior.

The post frames the benefit around workflow continuity: coding, iteration, and debugging cycles can move faster without requiring teams to switch to a smaller capability tier for day-to-day work.

What is explicitly claimed

• The claim is speed-oriented: 1.5x faster runtime in Codex when /fast is enabled.
• The post also claims no downgrade in intelligence and reasoning quality.
• The target use case is development velocity across iterative engineering tasks.

Why this is important for engineering teams

In practical software development, latency often compounds across many small agent interactions: planning, patching, running tools, and evaluating output. Even moderate speed gains can produce larger end-to-end savings over long sessions.

If the quality claim holds in production, /fast mode could reduce the common tradeoff between speed and reliability in coding assistants. Teams that already run Codex in CI-like loops or large refactoring tasks may see immediate throughput gains.

As with all vendor-declared performance updates, teams should validate on their own repositories and task mix before revising defaults.

Source: OpenAIDevs X post