GPT-5.4

2026-03-05 · Publisher : OpenAI · Domain : Artificial Intelligence

OpenAI launches GPT-5.4, a general-purpose reasoning model that integrates the coding capabilities of GPT-5.3-Codex and adds native computer use, with 1 million tokens of context.

What changes

  • First general-purpose OpenAI model with native computer use (computer tool, screenshots, mouse and keyboard).
  • Tool search in the API: tool definitions are loaded on demand, which reduced token usage by 47% on MCP Atlas at equal accuracy.
  • Native compaction for long tasks and a new "original" image detail level (up to 10.24 million pixels).
  • In ChatGPT, GPT-5.4 Thinking displays an upfront plan that can be edited while the response is in progress.
  • GPT-5.4 Pro for the hardest tasks.

### Specifications - Context: 1,050,000 tokens (272,000 standard, beyond that billed at 2x input and 1.5x output), maximum output 128,000 tokens. - Text and image input, text output. Knowledge cutoff: August 31, 2025. Effort: none (default) to xhigh. - Availability: ChatGPT (Plus, Team, Pro; Pro for GPT-5.4 Pro), API, Codex. - Identifiers: gpt-5.4, gpt-5.4-pro (Responses API only). ### Pricing | Model | Input | Cached input | Output | | --- | --- | --- | --- | | gpt-5.4 | $2.50 | $0.25 | $15.00 | | gpt-5.4-pro | $30.00 | n/a | $180.00 | Price per million tokens (USD). Batch and Flex at half price, Priority at 2x. ### Benchmarks | Evaluation | GPT-5.4 | GPT-5.4 Pro | GPT-5.3-Codex | GPT-5.2 | | --- | --- | --- | --- | --- | | GDPval (wins or ties) | 83.0% | 82.0% | 70.9% | 70.9% | | SWE-Bench Pro (public) | 57.7% | n/a | 56.8% | 55.6% | | Terminal-Bench 2.0 | 75.1% | n/a | 77.3% | 62.2% | | OSWorld-Verified | 75.0% | n/a | 74.0% | 47.3% | | BrowseComp | 82.7% | 89.3% | 77.3% | 65.8% | | MMMU Pro (no tools) | 81.2% | n/a | n/a | 79.5% | | Toolathlon | 54.6% | n/a | 51.9% | 45.7% | | MCP Atlas | 67.2% | n/a | n/a | 60.6% | Scores published by the vendor. ### Key takeaways GPT-5.4 merges coding, computer use and document work in a single model. The per-token price rises compared with GPT-5.2 ($2.50 versus $1.75 for input), offset according to OpenAI by better token efficiency.

Links

Tags : LLM, OpenAI, GPT-5, Raisonnement, Computer Use, Agent