The thesis

The tools for high-stakes work are either too general, shallow, or outdated.

Three things are true at once. Frontier models are racing horizontal: brilliant on general-purpose cases, but without domain structure, deep credential-gated sources, citation provenance, or built-in agentic capabilities that work independently of third-party subscriptions. On vertical work, these general models underperform their specialized cousins.

The legacy data terminals cost $12k - $250k+ a year, cover the ordinary course of business events and only a narrow set of extraordinary events (those that reach the newswires), are delivered through interfaces designed a decade ago, and have no agent runtime. And the one specialized player that has reached a multi-billion-dollar valuation has proven the vertical-AI thesis is right, while still leaving extraordinary events under-covered.

Quantimental builds vertical AI: products that own the ontology, the sources, the guardrails, and the workflow for a single domain, and are judged on outcomes inside that domain, not on general benchmarks. Vertical depth and provenance are the durable moat the horizontal race leaves behind.

Our first vertical

Marginal AI - extraordinary-event intelligence

Specialized intelligence for US public markets and prediction-market contracts. Agentic chat, alerts, and reports that monitor the extraordinary, non-operating events that fall outside the ordinary course of business (restatements, regulatory actions, litigation, restructuring, force majeure, market shocks, and more) and turn 1-2 analyst-days of manual triage into seconds, sourced and cited.

Marginal AI launches early June 2026 on web, iOS, Android, the Samsung Galaxy Store, and Huawei AppGallery. The same platform extends to further verticals; Garner AI is next.