OpenAI has launched GPT-5.4 with significant upgrades to visual understanding capabilities, now able to analyze images up to 10.24 million pixels or 6,000-pixel maximum dimensions. The company claims responses are 18 percent less likely to contain factual errors compared to previous versions. These improvements target enhanced knowledge-work capabilities for users across ChatGPT web and native apps, Codex, and the API.

This release comes amid competitive pressure from Anthropic, which recently expanded its memory features to free users and saw its largest single day of new sign-ups on March 2. OpenAI's 900 million user base reportedly faced some attrition after the company announced a Pentagon partnership, triggering a public feud with the Trump administration over military applications. Anthropic capitalized by rolling out previously subscriber-only features to all users.

GPT-5.4 is available today to Plus, Team, and Pro subscribers, who also gain access to GPT-5.4 Thinking. The Pro version is hitting the API, Edu, and Enterprise tiers. OpenAI needs to compete on both capability and cost efficiency to maintain its market position, making these visual and accuracy improvements strategically important for retaining users in an increasingly competitive AI landscape.