Today, OpenAI released their latest model, GPT-5.1. After extensive testing across a variety of Box AI use cases, we're seeing multiple areas of improvement, such as speed, accuracy, and conversationality, which will directly benefit Box AI users, especially those working with complex document understanding and structured data extraction.
Speed meets accuracy
Our evaluation reveals that GPT-5.1 delivers improvements on two critical dimensions: speed and accuracy. These aren't incremental gains. They represent fundamental leaps that transform what's possible with enterprise AI.
Latency improvements
GPT-5.1 achieves significantly lower latency across all enterprise document tasks. Latency measures the time between asking Box AI a question and receiving the first part of the response—essentially, your wait time. This includes time spent analyzing your query, retrieving relevant document sections, processing content through the AI model, and generating an answer. Lower latency means Box AI responds faster, making document interactions feel more conversational and immediate.
Our tests evaluated performance across different query types:
- Long documents, complex extraction: For straightforward questions on longer documents, latency drops from 45.6 to 16.7 seconds, enabling faster retrieval even on substantial content
- Long documents, analytical queries: Complex analytical questions require the model to synthesize and analyze information across large documents, and for these, processing time decreases from 19.3 to 9.1 seconds, enabling near-real-time analytical insights
- Long documents, multi-turn conversations: When users ask follow-up questions in an ongoing dialogue, GPT-5.1 maintains context from previous exchanges while processing new queries, and latency improves from 10.2 to 5.4 seconds, making interactive document exploration seamless
These speed gains mean your teams can interact with documents conversationally, ask follow-up questions, and get analytical insights without the frustrating delays that plagued earlier generations.
Accuracy gains on challenging extraction tasks
Beyond speed, GPT-5.1 demonstrates superior accuracy on the most challenging document extraction scenarios:
- Tabular data: Performance increases from 44% to 71%(a 61% relative improvement), making spreadsheet-like data extraction far more reliable
- Complex multi-field extraction: Success rates rise from 70% to 83% (an 18% relative improvement) when extracting numerous fields simultaneously
- Handwriting recognition: Improves from 38% to 42% (an 11% relative improvement) for one of AI's hardest challenges
- Long documents: Minimal change from 83% to 84% (a 2% relative improvement), showing GPT-5.1 maintains excellence on already-strong capabilities
What this means for Box customers
For organizations leveraging Box AI, this advancement translates to real business value.
Enhanced metadata extraction: Customers who rely on metadata extraction (MDE) across large document sets will see more accurate, consistent results, particularly when using taxonomy fields to ensure data quality and standardization. With 83% accuracy on complex multi-field extractions, you can trust automated processing for workflows that previously required human oversight.
Improved visual and structured data capture: Industries that work with forms, tables, invoices, and image-heavy documents will see transformative improvements. The 71% accuracy on tabular data means Box AI can now reliably extract information from the full range of enterprise content, not just clean text.
Interactive document analysis: With latency reductions of 50-70%, your teams can now have fluid, conversational interactions with documents. Ask a question, get an instant answer, drill deeper with follow-ups, and request analysis, all in real time.
Faster processing at scale: The speed improvements enable customers to process larger document volumes more efficiently. Whether you're analyzing thousands of contracts, processing daily invoice batches, or extracting metadata from archival documents, GPT-5.1 is a huge improvement.
Performance across the board
Our evals reveal strong performance across multiple dimensions:
- Higher accuracy: GPT-5.1 maintains high accuracy across diverse file types, field types, datasets, and industries
- Consistent performance: Whether processing invoices, contracts, forms, or unstructured documents, the model delivers reliable results
- Industry adaptability: From manufacturing and financial services to healthcare and retail, GPT-5.1 demonstrates robust performance across vertical-specific use cases
- Conversational: GPT-5.1 feels more intuitive and natural, following instructions more reliably and communicating in a warmer style with less jargon.
Real-world impact
The taxonomy and combined speed and accuracy improvements are particularly significant for customers managing high-volume document-processing workflows. Organizations that are using GPT-5.1 in Box AI can expect:
- Reduced manual review and correction time, especially for documents with tables, images, and complex layouts
- Higher confidence in automated metadata tagging, with 83% accuracy even on the most challenging multi-field extractions
- Better data quality for downstream analytics and reporting
- Improved compliance and governance through consistent categorization
- Interactive document exploration that feels responsive and natural
- The ability to scale AI-powered workflows to larger document volumes without proportional increases in processing time
- A more intuitive workflow-building experience, with GPT-5.1's improved instruction-following reducing the time spent refining prompts and templates
- Clearer, more natural outputs that communicate in a warmer style with less jargon, requiring less editing for both internal and customer-facing content
Get started with GPT-5.1 in Box AI
The advancements in GPT-5.1 represent a step toward more intelligent enterprise AI. The breakthrough in minimal reasoning translates to practical business outcomes: faster, more accurate document analysis and truly reliable structured data extraction. With GPT-5.1, businesses can implement AI solutions that process, analyze, and understand complex metadata extraction at scale, meeting the high standards required for mission-critical operations.
Ready to get started? GPT-5.1 is now available in the API and AI Studio under preview mode.
