Key Takeaways
- • Halluminate's Westworld Finance Diligence Bench found the best-performing frontier AI model configuration scored an average of just 51% when completing a full private-equity company due-diligence workflow across 88 tasks.
- • The benchmark uses tasks drawn from anonymized real-world private-equity transactions, written and reviewed by practicing deal professionals, with each deliverable graded on a 0-to-1 scale using expert-weighted checks.
- • Halluminate, a nine-employee company founded in 2024, closed a $30 million Series A led by Oak HC/FT on October 1, 2026, bringing total funding to $38.5 million.
- • Reports indicate Anthropic's Claude Opus 5 posted the highest mean score among seven tested configurations, with OpenAI's GPT-5.6 Sol roughly even while using fewer tokens; Halluminate has not published a complete model-by-model scoring table.
- • The results point to a copilot-first rather than autonomy-first adoption model in finance, with PE firms and banks likely to demand independent benchmark testing from AI vendors.
Summary
Halluminate released a benchmark showing that the strongest frontier AI model configuration completed an end-to-end private-equity due-diligence workflow with an average score of 51%. The Westworld Finance Diligence Bench comprises 88 tasks drawn from anonymized real-world private-equity transactions, written and reviewed by practicing deal professionals rather than academic researchers. The tasks cover financial modeling, quality-of-earnings analysis, document and diligence review, cross-document reasoning, and production of transaction deliverables, requiring agents to extract facts from a data room, flag inconsistencies, and produce work an investment committee could use.
Halluminate evaluated seven frontier AI model configurations but has not published a complete model-by-model scoring table. According to reports around the release, Anthropic's Claude Opus 5 posted the highest mean score, while OpenAI's GPT-5.6 Sol performed roughly even with it using meaningfully fewer tokens. xAI's Grok was described as the most token-efficient model when permitted to write and execute code. Halluminate has not disclosed the remaining four models or released exact individual percentages, so these details should be treated as directionally reported.
The company announced a $30 million Series A led by Oak HC/FT, a growth-equity firm focused on healthcare and fintech, with participation from Y Combinator, Orange Collective, FT Partners, and Heavybit. Total funding since the company's 2024 founding by CEO Jerry Wu and CTO Wyatt Marshall reached $38.5 million. According to Fortune's reporting, Halluminate counts four of the top U.S. AI labs among its customers, though these have not been named. Halluminate builds benchmarks and reinforcement-learning environments rather than competing frontier models.
The result highlights a recurring pattern in AI evaluation: models that perform well on short, isolated tasks degrade as workflows add steps, because errors compound across interdependent work. Practitioners cited in the coverage argue that a score near the midpoint creates false confidence, since a deal team may spot-check AI output and miss discrepancies that diligence exists to catch. The benchmark is expected to accelerate a copilot-first model in which AI handles first-pass extraction and drafting while human professionals retain judgment and sign-off.
The benchmark adds pressure on frontier labs to submit capability claims to independent verification before enterprise buyers rely on them. Halluminate operates alongside evaluators such as METR and Epoch AI, as well as data infrastructure providers like Scale AI and Mechanize, differentiating itself through a narrow, domain-specific focus graded by working deal professionals.
Halluminate evaluated seven frontier AI model configurations but has not published a complete model-by-model scoring table. According to reports around the release, Anthropic's Claude Opus 5 posted the highest mean score, while OpenAI's GPT-5.6 Sol performed roughly even with it using meaningfully fewer tokens. xAI's Grok was described as the most token-efficient model when permitted to write and execute code. Halluminate has not disclosed the remaining four models or released exact individual percentages, so these details should be treated as directionally reported.
The company announced a $30 million Series A led by Oak HC/FT, a growth-equity firm focused on healthcare and fintech, with participation from Y Combinator, Orange Collective, FT Partners, and Heavybit. Total funding since the company's 2024 founding by CEO Jerry Wu and CTO Wyatt Marshall reached $38.5 million. According to Fortune's reporting, Halluminate counts four of the top U.S. AI labs among its customers, though these have not been named. Halluminate builds benchmarks and reinforcement-learning environments rather than competing frontier models.
The result highlights a recurring pattern in AI evaluation: models that perform well on short, isolated tasks degrade as workflows add steps, because errors compound across interdependent work. Practitioners cited in the coverage argue that a score near the midpoint creates false confidence, since a deal team may spot-check AI output and miss discrepancies that diligence exists to catch. The benchmark is expected to accelerate a copilot-first model in which AI handles first-pass extraction and drafting while human professionals retain judgment and sign-off.
The benchmark adds pressure on frontier labs to submit capability claims to independent verification before enterprise buyers rely on them. Halluminate operates alongside evaluators such as METR and Epoch AI, as well as data infrastructure providers like Scale AI and Mechanize, differentiating itself through a narrow, domain-specific focus graded by working deal professionals.
Go deeper than the headline
Research every firm in this story, track the ones that matter and get the daily briefing — free for 7 days.