Earlier this week, OpenAI launched GPT-5.3 Instant, promising to make ChatGPT more natural while using its most popular model. Now, OpenAI has returned with GPT-5.4 Thinking and Pro, a general update to its most capable boundary model.
OpenAI last updated this model in December with GPT-5.2 "red alert" in response to competition from Google's Gemini.
As for what's new today, OpenAI highlights six major improvements in these areas:
- Coding, document understanding, tool usage, and instruction following
- Image recognition and multimodal tasks
- Long-term task execution and multi-step agent workflows
- Token efficiency and end-to-end performance in tool-heavy workloads
- Agentic web search and multi-source synthesis, especially for hard-to-find information
- Document-intensive and spreadsheet-intensive workflows in customer service, analytics, and finance
GPT-5.4 now offers a 1M token context window:
GPT-5.4 supports a 1M token context window, making it easier to analyze all codebases, long document collections, or extended agent paths in a single request.
OpenAI also notes that GPT-5.4 is the first "main model" with built-in computer usage:
GPT-5.4 is the first main model with built-in computer usage capabilities that can directly interact with software to complete, verify, and correct tasks.
Additionally, OpenAI states that GPT-5.4 is the first main model trained to support longer agent paths while maintaining core context.
OpenAI indicates that GPT-5.4 is being "rolled out gradually" in ChatGPT and Codex today. GPT-5.4 Thinking is available for Plus, Team, and Pro subscribers and will replace GPT-5.2 Thinking, which will expire in three months. Learn more about OpenAI's announcement here.
Do more with your iPhone
Apple AirTag 2 | Add Find My tracking to anything
Beats USB-A to USB-C Cable | Official CarPlay cable
Wireless CarPlay Adapter | No more plugging in
AirPods 4 | Apple's most affordable AirPods with ANC
Comments
(2 Comments)