Scaled Cognition Raises $100 Million Series A

Per a GlobeNewswire press release, Scaled Cognition closed a $100 million Series A round led by Khosla Ventures on June 25, 2026, with participation from Genesys, according to FinSMEs. The Wall Street Journal places the company valuation at about $750 million, with coverage also by The Next Web. The company says the funding will be used to expand its research team and accelerate enterprise deployments, per the GlobeNewswire announcement. The press release and multiple news reports describe the company as already running production deployments with Fortune 500 customers across financial services, healthcare, telecom, and insurance.

Per the GlobeNewswire release and reporting, Scaled Cognition's flagship model is APT (Agentic Pretrained Transformer). The press release characterizes APT as delivering conversational quality comparable to leading LLMs while reducing hallucinations, offering "policy-adherent performance," and being smaller, faster, and less expensive than frontier models. The release also describes the company platform as including agentic tooling, live agent monitoring, and simulation and evaluation frameworks, and notes availability for VPC and self-hosted deployment.

The Wall Street Journal quoted CEO Dan Roth describing frontier models as "amazing," but "sort of like schizophrenic geniuses," per coverage by The Next Web. The GlobeNewswire release quotes Roth: "We spent years trying to apply AI to business applications and found it was essentially impossible to make these systems reliable."

Per the company press release, APT is presented as an architecture-built approach to reliability rather than a safety wrapper layered on an existing frontier model. The release claims APT provides "Super-Reliable Intelligence," and is optimized for enterprise constraints such as lower compute cost and self-hosting options. News coverage by The Next Web and FinSMEs repeats those product and deployment claims as reported by the company. CTO and co-founder Dan Klein, a UC Berkeley professor of AI who previously co-founded and sold an agentic AI company to Microsoft, told The Next Web: "Reliability is engineered into the architecture of our models, not bolted on after the fact. If you want AI to take real actions on behalf of customers, that's the problem you have to solve."

Editorial analysis: Companies marketing "hallucination-free" models are responding to an acute enterprise barrier to adoption, especially in regulated verticals. Observed patterns in similar transitions show buyers and integrators prioritize architectures that enable verifiable correctness, audit trails, and on-prem or VPC deployment options, because external API dependence and opaque failure modes complicate compliance and remediation.

Editorial analysis: Financial services, healthcare, telecom, and insurance, the sectors named in the press release, have lower tolerance for unverifiable outputs. Vendors that combine monitoring, simulation-based evaluation, and smaller, auditable models can shorten procurement cycles compared with purely frontier-model dependent offerings.

Editorial analysis: Observers should track independent benchmarks and red-team results for APT, adoption case studies from the named verticals, and whether third-party audits confirm reduced hallucination rates. Also watch for technical disclosures from Scaled Cognition or peer-reviewed publications that explain the architectural differences underlying APT versus mainstream LLMs.

Editorial analysis: Funding of this size at Series A, led by a top-tier VC, increases the likelihood of rapid hiring and partnerships; practitioners should follow open-source artifacts, evaluation suites, or published APIs that would enable direct comparison with existing models.