DiligenceBench benchmarks equity-research agents, with Muse Spark 1.1 reaching 57.4%
karinanguyen · x · 2026-07-22
DiligenceBench benchmarks equity-research agents
A new benchmark, DiligenceBench, evaluates whether AI agents can do real public-equity research.
- The charted results show Meta Muse Spark 1.1 leading the finance harness at 57.4%, followed by GLM 5.2, Sonnet 4.6, and GPT-5.6 Sol.
- The authors found that stronger models benefit most from generic tools that unlock execution, while weaker models need more opinionated, domain-specific scaffolding.
- For Inkling, a generic sandbox barely helped (20.9% → 22.5%), but the finance harness raised it to 32.8%, with the biggest gain coming from factual accuracy.
- The benchmark also changes the price–performance frontier: most models get both better and cheaper under the finance harness, with GLM 5.2 leading on absolute performance and MiniMax M3 appearing among the efficiency winners.
More from Venture
- Analyst: Ant has the most fungible AI workload of any model provider today — BenBajarin · 2026-10-26
- Anthropic S-1: training infra edge cut per-task cost 40% and doubled revenue growth — BenBajarin · 2026-10-26
- Anti-AI content is algorithm gold and very profitable, quips commentator — annetgriffin · 2026-10-26
- SaaS hype is gone: the shift to AI-native services that sell outputs, not software — mitchdeg · 2026-10-26
- Anthropic S-1 takeaway: it's good to be a hyperscaler — matt_slotnick · 2026-10-26
- SpaceX S-1 cites $28.5T TAM; Anthropic points to $40T cognitive-labor market — annbordetsky · 2026-10-26