Sequoia Leads $200M Series B for Anthropic Competitor Focused on Verifiable Reasoning Models
Tags AI · VC · Enterprise · Infrastructure
Sequoia Capital led a $200M Series B at a $2.8B valuation for Cognition Labs, a startup building AI models that produce mathematically verifiable reasoning traces. The company's flagship model, "Prover-72B," generates step-by-step proofs for mathematical and logical claims that can be mechanically checked by a Lean 4 verifier. Unlike standard RLHF, the training objective directly optimizes for verifiable correctness rather than human preference. The round includes participation from Andreessen Horowitz, Lightspeed, and Nvidia's NVentures. Cognition plans to launch an API for formal verification-assisted coding by Q4 2026.
Technical significance
This represents a shift from "vibes-based" alignment (RLHF) to formally verifiable AI reasoning — a critical requirement for high-stakes domains like theorem proving, smart contract auditing, safety-critical code generation, and scientific discovery. The $2.8B valuation for a pre-revenue company signals investor conviction that verifiable reasoning will be a defensible moat. For the ecosystem, it creates pressure on frontier labs to adopt proof-carrying outputs, potentially standardizing Lean/Coq integration in model APIs within 12-18 months.