The Frontier Eval Gap: Why Leaderboard Wins Don't Survive M&A IntegrationLa brecha de evals en la frontera: por qué las victorias en leaderboards no sobreviven la integración M&A

July 2026 exposes a structural problem acquirers ignore: the same model generation that wins public coding leaderboards fails real enterprise workflows. Kimi K3 scores 1,679 on Arena Frontend Code while Vals AI's Finance Agent v2 shows frontier models peaking below 58% on multi-step analyst tasks — planning, retrieval, calculation, and synthesis on actual financiero documents.

Post-close integration fails when compradores assumed benchmark leadership transfers to customer workflows. Three failure modes recur. Domain mismatch: coding-optimized models underperform on structured business data (the gap Prior Labs' TabPFN explicitly targets). Harness dependency: Arena scores reflect prompt engineering, tool access, and sandbox setup — assets that may not transfer with the model weights. Regression risk: model updates can erase leaderboard advantages dentro de a quarter; earnouts tied to static benchmarks create adversarial incentives.

Pricing fix: discount projected AI ingresos where claims rely on public benchmarks not reproduced in comprador-controlled evals; require holdbacks for targets whose moat is model performance sin proprietary data or workflow lock-in.

My View on This

the eval gap is the AI analogue of ingresos quality diligence — separate headline metrics from durable capability. antes de signing, run 30 tasks from the objetivo's top three customer segments on the adquirente's infrastructure. If scores diverge more than 15 points from vendor claims, the synergy case is unproven and the multiple debe reflect software-plus-services, not frontier IP.

Fuentes

• Vals AI, Finance Agent v2 benchmark
• Tech Startups, Kimi K3 Arena
• SAP News, Prior Labs structured data