ToltIQ recently adopted the CorpFin V2 private validation benchmark, created by Vals AI, to measure how the leading LLM’s that are leveraged through our due diligence platform performs against standalone OpenAI and Anthropic models. The results demonstrate significant advantages for purpose-built systems over generic LLMs in private equity workflows.
We selected this benchmark because it offers independent, third-party validation with objective evaluation criteria. The CorpFin V2 benchmark focuses specifically on corporate finance tasks relevant to PE due diligence, and the private validation set reduces the risk of data contamination from model training.
Maya Boeye - Head of AI Research (Provided conceptual guidance, supervised the research process and report review.)
Steiner Williams - Private Equity AI Researcher (Led the benchmarking exercise, data analysis, and reporting.)
Using the Vals AI CorpFin V2 benchmark, we measured the accuracy of LLM’s leveraged through ToltIQ's platform against standalone implementations of GPT-5.1, GPT-5.2, Claude Sonnet 4.5, and Claude Opus 4.5.
The benchmarking results showed that using the LLMs within the ToltIQ platform outperformed standalone models across every configuration:
The CorpFin benchmark evaluates AI models on 360 corporate finance questions derived from 18 documents, measuring real-world corporate financial analysis capabilities. We implemented a controlled methodology using restricted chat sessions for each benchmark document, mirroring production usage. All responses were scored using Vals AI's standardized evaluator, supplemented by manual review where automated evaluation proved insufficient. We believe human oversight remains essential for capturing nuances.
Vals AI is an independent AI evaluation company that publishes their own benchmark testing for finance, legal, and other professional domains. The standalone LLM scores above are the official results reported by Vals AI for the CorpFin benchmark. The standalone model scores above are the official standalone model scores reported by Vals AI for the CorpFin benchmark.
To establish statistical significance, each prompt was executed five times per model, yielding 14,400 total evaluated responses across the ToltIQ platform and standalone models.
The CorpFin benchmark drove a concrete platform decision. Due to the performance differential with GPT-5.1 scoring 80.5% compared to GPT-5.2's 71.5%, we're reverting back to GPT-5.1 as our default OpenAI model.
It’s worth noting, however, that while Claude Opus 4.5 achieved the highest accuracy today, and GPT-5.2 performed poorly on a relative basis, the competitive landscape shifts constantly, with providers leapfrogging each other release after release. Our model-agnostic approach means users always benefit from ToltIQ's platform enhancements, which consistently boost base model accuracy for PE due diligence tasks beyond what standalone versions achieve.
ToltIQ achieves these meaningful enhancements through its purpose-built RAG systems, specialized retrieval methods optimized for financial documents, and verification layers that standalone ChatGPT and Claude implementations don't provide. This architecture advantage translates directly into more accurate due diligence outputs.
Our research team has systematically divided the benchmark into private equity due diligence use cases for performance analysis of LLMs both within and outside of our platform. We’ll be sharing a long form analysis of the comprehensive results with details on specific workflows and performance in the coming weeks.
If you’re interested in learning more about this research or would like to see ToltIQ in action, please click here to request a demo, and we’ll reach out to schedule a convenient time.
© 2026 ToltIQ
