GPT 5.4 includes new extreme reasoning mode and 1M context, details below
The new model can run tasks for hours with lower error rates, designed for agents and automation.
OpenAI has unveiled GPT-5.4, a significant update to its flagship model series, as reported by The Information. The new version introduces two headline features: a massive 1 million token context window and a novel 'Extreme reasoning' mode. This mode dedicates substantially more computational resources to complex problem-solving, allowing the model to engage in deeper, more sustained thinking. The update represents a strategic move to achieve parity with long-context offerings from rivals like Google's Gemini and Anthropic's Claude, while also marking a shift towards a more aggressive, monthly update cadence for OpenAI's models.
The technical enhancements are geared towards enterprise and research applications. The 1M token context and improved memory architecture enable the model to handle 'long-horizon' tasks—such as intricate data analysis or multi-stage coding projects—that can run continuously for hours. This is coupled with lower error rates in complex operations, making it particularly suitable for powering autonomous agents and automation workflows (like an advanced Codex). The focus on scientific research and complex problem-solving indicates OpenAI is targeting high-stakes, professional use cases where reasoning depth and reliability are critical, moving beyond conversational AI into core operational infrastructure.
- Introduces a 1 million token context window, matching competitor capabilities for processing massive documents.
- New 'Extreme reasoning' mode allocates more compute for complex tasks, enabling multi-hour analysis with lower error rates.
- Explicitly designed for autonomous agents and automation, signaling a push into enterprise workflow integration.
Why It Matters
Enables AI to tackle sustained, complex scientific and operational problems, moving from assistant to autonomous agent.