There is no such thing as a scarcity of AI benchmarks available in the market at the moment, with standard choices like Humanity's Final Examination (HLE), ARC-AGI-2 and GDPval, amongst quite a few others.
AI brokers excel at fixing summary math issues and passing PhD-level exams that the majority benchmarks are based mostly on, however Databricks has a query for the enterprise: Can they really deal with the document-heavy work most enterprises want them to do?
The reply, in line with new analysis from the information and AI platform firm, is sobering. Even the best-performing AI brokers obtain lower than 45% accuracy on duties that mirror actual enterprise workloads, exposing a crucial hole between educational benchmarks and enterprise actuality.
"If we focus our research efforts on getting better at [existing benchmarks], then we're probably not solving the right problems to make Databricks a better platform," Erich Elsen, principal analysis scientist at Databricks, defined to VentureBeat. "So that's why we were looking around. How do we create a benchmark that, if we get better at it, we're actually getting better at solving the problems that our customers have?"
The result’s OfficeQA, a benchmark designed to check AI brokers on grounded reasoning: Answering questions based mostly on complicated proprietary datasets containing unstructured paperwork and tabular information. In contrast to current benchmarks that target summary capabilities, OfficeQA proxies for the economically worthwhile duties enterprises really carry out.
Why educational benchmarks miss the enterprise mark
There are quite a few shortcomings of standard AI benchmarks from an enterprise perspective, in line with Elsen.
HLE options questions requiring PhD-level experience throughout various fields. ARC-AGI evaluates summary reasoning by visible manipulation of coloured grids. Each push the frontiers of AI capabilities, however don't mirror day by day enterprise work. Even GDPval, which was particularly created to judge economically helpful duties, misses the goal.
"We come from a pretty heavy science or engineering background, and sometimes we create evals that reflect that," Elsen mentioned. " So they're either extremely math-heavy, which is a great, useful task, but advancing the frontiers of human mathematics is not what customers are trying to do with Databricks."
Whereas AI is often used for buyer help and coding apps, Databricks' buyer base has a broader set of necessities. Elsen famous that answering questions on paperwork or corpora of paperwork is a typical enterprise job. These require parsing complicated tables with nested headers, retrieving info throughout dozens or a whole bunch of paperwork and performing calculations the place a single-digit error can cascade into organizations making incorrect enterprise choices.
Constructing a benchmark that mirrors enterprise doc complexity
To create a significant check of grounded reasoning capabilities, Databricks wanted a dataset that approximates the messy actuality of proprietary enterprise doc corpora, whereas remaining freely obtainable for analysis. The crew landed on U.S. Treasury Bulletins, printed month-to-month for 5 many years starting in 1939 and quarterly thereafter.
The Treasury Bulletins examine each field for enterprise doc complexity. Every bulletin runs 100 to 200 pages and consists of prose, complicated tables, charts and figures describing Treasury operations: The place federal cash got here from, the place it went and the way it financed authorities operations. The corpus spans roughly 89,000 pages throughout eight many years. Till 1996, the bulletins have been scans of bodily paperwork; afterwards, they have been digitally produced PDFs. USAFacts, a company whose mission is "to make government data easier to access and understand," partnered with Databricks to develop the benchmark, figuring out Treasury Bulletins as splendid and guaranteeing questions mirrored sensible use circumstances.
The 246 questions require brokers to deal with messy, real-world doc challenges: Scanned photos, hierarchical desk constructions, temporal information spanning a number of experiences and the necessity for exterior information like inflation changes. Questions vary from easy worth lookups to multi-step evaluation requiring statistical calculations and cross-year comparisons.
To make sure the benchmark requires precise document-grounded retrieval, Databricks filtered out questions that LLMs may reply utilizing parametric information or internet search alone. This eliminated easier questions and a few surprisingly complicated ones the place fashions leveraged historic monetary data memorized throughout pre-training.
Each query has a validated floor reality reply (usually a quantity, generally dates or small lists), enabling automated analysis with out human judging. This design selection issues: It permits reinforcement studying (RL) approaches that require verifiable rewards, just like how fashions practice on coding issues.
Present efficiency exposes elementary gaps
Databricks examined Claude Opus 4.5 Agent (utilizing Claude's SDK) and GPT-5.1 Agent (utilizing OpenAI's File Search API). The outcomes ought to give pause to any enterprise betting closely on present agent capabilities.
When supplied with uncooked PDF paperwork:
Claude Opus 4.5 Agent (with default pondering=excessive) achieved 37.4% accuracy.
GPT-5.1 Agent (with reasoning_effort=excessive) achieved 43.5% accuracy.
Nonetheless, efficiency improved noticeably when supplied with pre-parsed variations of pages utilizing Databricks' ai_parse_document, indicating that the poor uncooked PDF efficiency stems from LLM APIs combating parsing moderately than reasoning. Even with parsed paperwork, the experiments present room for enchancment.
When supplied with paperwork parsed utilizing Databricks' ai_parse_document:
Claude Opus 4.5 Agent achieved 67.8% accuracy (a +30.4 share level enchancment)
GPT-5.1 Agent achieved a 52.8% accuracy (a +9.3 share level enchancment)
Three findings that matter for enterprise deployments
The testing recognized crucial insights for practitioners:
Parsing stays the basic blocker: Complicated tables with nested headers, merged cells and strange formatting ceaselessly produce misaligned values. Even when given precise oracle pages, brokers struggled primarily as a consequence of parsing errors, though efficiency roughly doubled with pre-parsed paperwork.
Doc versioning creates ambiguity: Monetary and regulatory paperwork get revised and reissued, that means a number of legitimate solutions exist relying on the publication date. Brokers typically cease looking out as soon as they discover a believable reply, lacking extra authoritative sources.
Visible reasoning is a spot: About 3% of questions require chart or graph interpretation, the place present brokers persistently fail. For enterprises the place information visualizations talk crucial insights, this represents a significant functionality limitation.
How enterprises can use OfficeQA
The benchmark's design permits particular enchancment paths past easy scoring.
"Since you're able to look at the right answer, it's easy to tell if the error is coming from parsing," Elsen defined.
This automated analysis permits fast iteration on parsing pipelines. The verified floor reality solutions additionally allow RL coaching just like coding benchmarks, since there's no human judgment required.
Elsen mentioned the benchmark gives "a really strong feedback signal" for builders engaged on search options. Nonetheless, he cautioned in opposition to treating it as coaching information.
"At least in my imagination, the goal of releasing this is more as an eval and not as a source of raw training data," he mentioned. "If you tune too specifically into this environment, then it's not clear how generalizable your agent results would be."
What this implies for enterprise AI deployments
For enterprises at present deploying or planning document-heavy AI agent techniques, OfficeQA gives a sobering actuality examine. Even the newest frontier fashions obtain solely 43% accuracy on unprocessed PDFs and fall in need of 70% accuracy even with optimum doc parsing. Efficiency on the toughest questions plateaus at 40%, indicating substantial room for enchancment.
Three instant implications:
Consider your doc complexity: In case your paperwork resemble the complexity profile of Treasury Bulletins (scanned photos, nested desk constructions, cross-document references), anticipate accuracy effectively under vendor advertising and marketing claims. Take a look at in your precise paperwork earlier than manufacturing deployment.
Plan for the parsing bottleneck: The check outcomes point out that parsing stays a elementary blocker. Price range time and assets for customized parsing options moderately than assuming off-the-shelf OCR will suffice.
Plan for laborious query failure modes: Even with optimum parsing, brokers plateau at 40% on complicated multi-step questions. For mission-critical doc workflows that require multi-document evaluation, statistical calculations or visible reasoning, present agent capabilities might not be prepared with out vital human oversight.
For enterprises trying to lead in AI-powered doc intelligence, this benchmark gives a concrete analysis framework and identifies particular functionality gaps that want fixing.
