OpenAI Ships GPT-5.4 With 1M Token Context and New Reasoning Variants
OpenAI has released GPT-5.4, its latest frontier model designed for professional work. The release introduces three variants: a standard model, GPT-5.4 Thinking (a reasoning-focused model), and GPT-5.4 Pro (optimized for maximum performance). The model is now available across ChatGPT and the OpenAI API.
Key Capabilities
GPT-5.4 supports up to 1 million tokens via the API, making it OpenAI's longest-context model to date. The release also includes Tool Search, which allows models to look up tool definitions dynamically rather than including all definitions in system prompts, reducing token consumption and costs for complex agent systems.
OpenAI claims the model is 33% less likely to make errors in individual claims compared to GPT-5.2, with overall responses 18% less likely to contain errors. The model demonstrates significantly improved token efficiency, solving the same problems with substantially fewer tokens than its predecessor.
Benchmark Results
GPT-5.4 achieved state-of-the-art results across multiple evaluations:
- OSWorld-Verified (desktop navigation): 75.0% success rate, surpassing human performance at 72.4%
- ARC-AGI-2 (visual/abstract reasoning): 73.3% vs. GPT-5.2's 52.9%
- FrontierMath (advanced math): 27.1% vs. 14.6%
- GDPval (economic knowledge): 83% score
Pricing and Availability
The API pricing starts at $2.50 per million input tokens and $15 per million output tokens for the standard tier. The Pro tier runs $30 input and $180 output per million tokens. In ChatGPT, GPT-5.4 Thinking is available to Plus, Team, and Pro subscribers, with GPT-5.2 Thinking set to retire on June 5, 2026.