OpenAI Surprises with GPT-5.4 Release: 1M Token Window and Office-Focused Tools
OpenAI Surprises with GPT-5.4 Release: 1M Token Window and Office-Focused Tools
OpenAI has just announced GPT-5.4, a frontier model heavily focused on professional productivity. The major update brings an incredible context window of over 1 million tokens, advanced logical reasoning modes (the “Thinking” version), and a brand-new native computer use feature.
Contexto
Just one day after the release of GPT-5.3 Instant, OpenAI accelerated its release schedule. This swift move, described as arriving “sooner than you think,” occurs amid fierce competition in the AI market against giants like Anthropic (Claude) and Google (Gemini). It also serves as a strategic response to a recent user exodus following controversies over the company’s military contracts. It is clear that OpenAI is now aggressively targeting the enterprise market to consolidate its dominance.
Detalhes
GPT-5.4 introduces a robust set of improvements and new features, designed to minimize errors and maximize efficiency in workplace tasks:
- Expanded Context Window: The new model breaks the 1 million token mark, more than double the 400,000 token capacity of GPT-5.2. There are even leaks pointing to 2 million token support, allowing the processing of entire databases, books, or huge codebases in a single prompt.
- “Pro” and “Thinking” Versions: In addition to the standard version, GPT-5.4 introduces a new reasoning model called “GPT-5.4 Thinking”, promising an “extreme” level of logic to handle the most complex analytical and mathematical tasks.
- Computer Use Mode and Financial Plugins: OpenAI introduced a “native computer use” mode, allowing the model to interact directly with user interfaces. Alongside this, native financial plugins for Microsoft Excel and Google Sheets are now available, automating heavy corporate workflows.
- Renewed Efficiency: The company claims that GPT-5.4 is much more efficient, using up to 47% fewer tokens on specific tasks compared to its predecessors, making the model faster and potentially cheaper for the end user.
- Full-Resolution Vision: Leaked commits in the Codex repository indicate dramatic improvements in how the model handles images, requiring the new version for advanced visual manipulation features.