OpenAI has launched GPT-5.4, its latest frontier model positioned as the most capable and efficient for professional work. It merges coding capabilities from GPT-5.3-Codex with improved support for spreadsheets, documents, and presentations. Key improvements include better web search, longer context handling, enhanced computer-use capabilities, and a 33% reduction in false claims compared to GPT-5.2. A notable new feature lets GPT-5.4 Thinking display an upfront reasoning plan, enabling users to steer the model mid-response. The one-million-token context window exits beta for API users. The model comes in Thinking and Pro tiers; Pro is priced at $30/$180 per million input/output tokens, making it OpenAI's most expensive model yet, though OpenAI claims greater token efficiency offsets costs. On benchmarks, GPT-5.4 outperforms predecessors, Anthropic's Opus 4.6, and Google's Gemini 3.1 Pro across coding, agentic, and knowledge-work tasks. Rollout is underway across ChatGPT and the Codex app.
Table of contents
More expensive models, but less token usageBenchmarksBetter at knowledge workAvailabilitySort: