Visual Enhancements in GPT-5.4

OpenAI’s latest model, GPT-5.4, brings notable improvements to visual understanding. The system can now carefully analyze images containing up to 10.24 million pixels and supports a maximum image dimension of 6,000 pixels. These upgrades expand the range of visual tasks the model can handle, offering users more detailed image analysis capabilities.

Boosted Accuracy Claims

Alongside visual upgrades, OpenAI asserts that GPT-5.4’s responses are 18 percent less likely to contain factual errors than previous versions. This reduction in error rate is presented as a step toward more reliable and trustworthy AI-generated content.

Competitive Dynamics with Anthropic

The launch occurs amid shifting user dynamics. After OpenAI announced a deal with the Pentagon, some users reportedly moved to Anthropic’s platform. In response, Anthropic made its memory feature, previously limited to subscribers, available to free users and introduced a tool for importing memory from other sources. Anthropic highlighted March 2 as its largest single day for new sign‑ups, underscoring the competitive pressure.

Strategic Focus on Capability, Cost, and Efficiency

OpenAI emphasizes the need to balance capability with cost and token efficiency to retain its large user base. By enhancing visual performance and reducing factual errors, the company aims to sustain its popularity while addressing competitive challenges.

Availability Across Platforms and Plans

GPT-5.4 is now accessible through the ChatGPT web interface, native applications, Codex, and the API. Subscribers to Plus, Team, and Pro tiers receive a version labeled GPT-5.4 Thinking, while GPT-5.4 Pro is rolled out to API, Edu, and Enterprise customers. This tiered deployment ensures that a broad spectrum of users can benefit from the new model’s capabilities.

Used: News Factory APP - news discovery and automation - ChatGPT for Business