Today, with the release of OpenAI’s GPT-5.2, we are seeing a shift to models becoming more specialized in certain knowledge or tasks. We evaluated this new model using our updated Box AI Enterprise Eval, and the data reveals a critical evolution: GPT-5.2 has achieved a leap in both speed and complex reasoning.
Speeding up the subject-matter expert
Deep expertise usually requires time, but GPT-5.2 defies this rule. At the same time it became smarter about specialized topics, it also got faster.
We benchmarked GPT-5.2 with medium reasoning against its predecessors, GPT-5 and GPT-5.1, on medium reasoning and focusing on use cases involving long contexts. The latency reductions, measured in time to first token, are stark, showing that complex workflows won’t stall while waiting for an answer.
- Complex extraction: Latency improved dramatically, dropping from 46s (GPT-5) and 17s (GPT-5.1) to just 12s with GPT-5.2
- Analytical queries: We saw the time to insight fall from 19s in GPT-5 and 9s in GPT-5.1 to a lightning-fast 7s
- Multi-turn queries: For conversational workflows, the model achieved a response time of 5s, compared to 10s for GPT-5 and 5s for GPT-5.1
Reasoning where it matters most
Our Box AI benchmark, focused on complex task automation and designed to be difficult, has further matured to more deeply measure model performance. In our latest testing on our updated evaluation sets, the average reasoning score across all industries bumped up from 59% on GPT-5.1 with medium reasoning to 66% on GPT-5.2 with medium reasoning. However, when we isolated specific, complex verticals, GPT-5.2 demonstrated a big shift in capability.
- Media and entertainment: GPT-5.1 scored 76%, but GPT-5.2 captured a jump to 81%
- Financial services: Reasoning capabilities rose from 67% for GPT-5.1 to 71% with GPT-5.2
- Life sciences and healthcare: Accuracy improved from 57% for GPT 5.1 to 59% with GPT 5.2
Transforming knowledge work: Real-world use cases
The evolution of GPT-5.2 allows us to deploy Box AI into more sophisticated, high-stakes use cases than ever before. Here’s how these improvements translate to tangible value in key industries.
1. Media and entertainment: Data-driven box office and distribution strategy
In M&E, release windows are shrinking while expectations for data-driven decisions are soaring. Studio teams juggle spreadsheets, territory reports, and franchise dashboards to understand how different markets fuel a film’s performance.
The challenge: A studio team needs to see which of its top releases rely most on North America vs. international audiences, based on global lifetime gross and regional earnings across dozens of box office reports and franchise overviews.
The Box AI solution: Using GPT-5.2, Box AI ingests the folder of spreadsheets and decks, computes domestic and international share for each title, ranks the films by reliance on each market, and highlights franchise-level patterns, all in seconds (instead of hours of manual spreadsheet work).
2. Financial services: Accelerating investment due diligence
Speed and accuracy are the currency of financial services, where analysts must synthesize reports, analyze data, and pay attention to market news to make high-value decisions.
The challenge: A commercial bank’s credit team needs to compare a portfolio of corporate term loans and revolving credit facilities across dozens of credit agreements, amendments, and term sheets. They must line up pricing grids, covenants, collateral, maturities, and risk metrics to see which borrowers are underpriced, which structures are off-market, and where to focus repricing or renewal conversations.
The Box AI solution: Using GPT-5.2, Box AI ingests the full set of loan documents, normalizes key terms, and builds a side-by-side comparison: extracting rates and spreads, leverage and coverage covenants, collateral packages, financial reporting requirements, and key risk clauses. It then highlights outliers and concentration risks, surfaces which loans are most aggressively or conservatively structured, and produces a concise summary that turns days of manual document review into minutes of portfolio-level insight.
3. Life sciences and healthcare: Synthesizing complex clinical data
In an industry where every minute counts, researchers are often buried under mountains of unstructured data, trial notes, and regulatory submissions.
The challenge: A medical affairs team needs to synthesize findings from dozens of peer‑reviewed papers, conference abstracts, and regulatory summaries to understand safety and efficacy signals for a specific therapy across multiple subpopulations.
The Box AI solution: Using GPT-5.2, Box AI ingests the full corpus of PDFs and slide decks, identifies relevant endpoints and patient cohorts, extracts key outcomes and adverse events, and reconciles conflicting findings. It then produces a structured evidence table and a narrative summary of trends and gaps, turning weeks of manual literature review into minutes of high-confidence insight.
Experience the difference
GPT-5.2 is available today for all Box AI customers. We encourage you to test it not just on your simple documents, but on your hardest, most specialized content. Access GPT-5.2 in Box AI Studio and via the Box AI APIs.
