Editorial illustration for OpenAI launches GPT-5.4 with computer-use, Excel plugins, 17% BrowseComp boost
GPT-5.4 Transforms Finance with Excel Plugin Magic
OpenAI launches GPT-5.4 with computer-use, Excel plugins, 17% BrowseComp boost
OpenAI’s latest rollout, GPT‑5.4, adds a native “computer‑use” mode and plugs straight into Microsoft Excel and Google Sheets, promising a more hands‑on assistant for finance teams. The upgrade isn’t just a feature list; OpenAI is backing it with benchmark scores that aim to show how the model handles real‑world tasks. One of those tests, BrowseComp, gauges an agent’s ability to keep searching the web for obscure facts without losing track.
According to the company, GPT‑5.4 nudges its score up by 17 percentage points compared with GPT‑5.2, while the Pro variant hits an 89.3 percent mark that OpenAI calls a new state of the art. Another metric, OSWorld‑Verified, looks at how well the system navigates a desktop environment. These numbers are meant to give developers a concrete sense of progress before they start building on the new plugins.
On BrowseComp, which measures how well AI agents can persistently browse the web to find hard-to-locate information, OpenAI reports GPT-5.4 improving by 17% absolute over GPT-5.2, and GPT-5.4 Pro reaching 89.3%, described as a new state of the art. On OSWorld-Verified, which measures desktop navigat
On BrowseComp, which measures how well AI agents can persistently browse the web to find hard-to-locate information, OpenAI reports GPT-5.4 improving by 17% absolute over GPT-5.2, and GPT-5.4 Pro reaching 89.3%, described as a new state of the art. On OSWorld-Verified, which measures desktop navigation using screenshots plus keyboard and mouse actions, OpenAI reports GPT-5.4 at 75.0% success, compared to 47.3% for GPT-5.2, and notes reported human performance at 72.4%. On WebArena-Verified, GPT-5.4 reaches 67.3% success using both DOM- and screenshot-driven interaction, compared to 65.4% for GPT-5.2.
On Online-Mind2Web, OpenAI reports 92.8% success using screenshot-based observations alone. OpenAI also links computer use to improvements in vision and document handling. On MMMU-Pro, GPT-5.4 reaches 81.2% success without tool use, compared with 79.5% for GPT-5.2, and OpenAI says it achieves that result using a fraction of the "thinking tokens." On OmniDocBench, GPT-5.4's average error is reported at 0.109, improved from 0.140 for GPT-5.2.
What we now have is GPT‑5.4, released just two days after the GPT‑5.3 Instant update. Two flavors arrive: Thinking for general use and Pro for the most demanding tasks. Both are being rolled out through OpenAI’s paid API and the Codex development platform, while the Thinking variant will also be accessible elsewhere.
On the BrowseComp benchmark, which gauges an agent’s ability to keep searching the web for hard‑to‑find data, OpenAI reports a 17‑point absolute gain over GPT‑5.2, and the Pro model hits 89.3 %, a figure it calls state‑of‑the‑art. OSWorld‑Verified, a test of desktop navigation, is also mentioned, though the summary stops short of giving its results. The new Excel and Google Sheets plugins suggest a push toward tighter office‑software integration, yet how much these additions will change real‑world workflows remains unclear.
Likewise, the practical impact of the reported BrowseComp jump will depend on how developers employ the API. For now, the rollout adds more options, but the broader utility of GPT‑5.4’s enhancements is still to be demonstrated.
Further Reading
- Papers with Code - Latest NLP Research - Papers with Code
- Hugging Face Daily Papers - Hugging Face
- ArXiv CS.CL (Computation and Language) - ArXiv
Related Reading
Browse all OpenAI news →
Business & Startups OpenAI, a Series F San Francisco startup founded in 2015 by eight pioneers
Business & Startups Terminal-Bench 2.0 launches with Harbor, testing any container-installable agent
Business & Startups Zuckerberg Unveils Meta Compute to Build Global AI Infrastructure
Related Topic Gen AI app sessions up fivefold, downloads jump 778% as ChatGPT leads traffic
LLMs & Generative AI
Related Topic GPT-5 helps mathematicians offload tedious tasks, says Timothy Gowers
Open Source
From Archives Apple Music introduces optional AI labels to boost transparency
From Archives Anthropic CEO Dario Amodei returns to Pentagon talks to salvage deal
From Archives OpenAI launches GPT-5.4 and ChatGPT Agent, enabling computer‑task automation
Industry Applications
From Archives Amodei slams OpenAI in memo, urges automated audit‑ready evidence collection
Common Questions Answered
How does GPT-5.4 improve web browsing capabilities compared to previous versions?
OpenAI reports that GPT-5.4 significantly enhances web browsing performance on the BrowseComp benchmark, improving by 17% absolute over GPT-5.2. The Pro version reaches an impressive 89.3% success rate, which the company describes as a new state of the art in persistent web information retrieval.
What new features does GPT-5.4 introduce for productivity tools?
GPT-5.4 adds a native 'computer-use' mode and introduces direct plugins for Microsoft Excel and Google Sheets, making it a more hands-on assistant for finance teams. The model also demonstrates improved desktop navigation, achieving a 75.0% success rate on OSWorld-Verified tests, compared to 47.3% for the previous version.
What are the different versions of GPT-5.4 being released?
OpenAI is launching two variants of GPT-5.4: a Thinking version for general use and a Pro version designed for the most demanding tasks. These will be distributed through OpenAI's paid API and the Codex development platform, with the Thinking variant also being made available through additional channels.