A new family of models that cuts enterprise document AI cost by 8.9x while improving accuracy.
MOUNTAIN VIEW, Calif., April 30, 2026: AISquared, a leading provider of enterprise AI infrastructure, today announced ‘Bolt’, a family of purpose-built AI models that process enterprise documents at one-twentieth the cost of frontier alternative large language models like Claude Opus 4.6 while delivering task-specific accuracy. At one million invoices per month, Bolt reduces annual operating costs by an estimated $1.89 million.
Bolt addresses the “token burden” by lowering the unnecessary cost of using expensive frontier AI models for routine enterprise AI workloads, including document processing, retrieval, governance checks, and routing. By deploying purpose-built, specialized, smaller models within AISquared’s UNIFI infrastructure platform, enterprises can replace expensive monolithic deployments with a coordinated portfolio of efficiency models that can be run on-premise or in the cloud.
In benchmark evaluations, Bolt demonstrated that a smaller, specialized model designed for enterprise activities can outperform larger general-purpose alternatives on targeted enterprise tasks. In invoice parsing benchmarks, Bolt-VL-9B achieved stronger document extraction performance than larger foundation models by 7.2x while reducing the infrastructure cost of performance by 42%.
Separately, AISquared benchmarked a Bolt Instruct 32B-powered routing layer against a monolithic GPT-5.5 deployment on a 620-request mixed workload, reducing operating cost by 49% and improving weighted average latency by 42%. The routed setup shifted 58% of requests to smaller, lower-cost models while reserving higher-capability inference for the most demanding tasks.
The model also showed a strong safety and compliance profile, achieving a near-perfect PII detection score and leading among tested models on evaluations for content sensitivity.
“The next phase of enterprise AI will not be won by using the biggest model for every task,” said Darren Kimura, CEO and president of AISquared. “It will be won by using the right model for the right workflow. Bolt is designed for that reality. It reduces the token and infrastructure burden of AI while improving performance on the tasks enterprises actually need to run in production.”
Bolt reflects a broader shift in enterprise AI from model access to model efficiency. As organizations scale from pilots to production, the winning architecture is no longer one large model serving every task, but a portfolio of specialized models orchestrated through a governed enterprise AI infrastructure layer.
“Enterprises don’t need one giant model for every job,” said Dr. Jacob Renn, co-founder and chief data scientist at AISquared. “They need the right model for each task running in controlled environments with full visibility and tight integration into their data systems.”
Bolt strengthens AISquared’s position as a provider of both specialized AI models and enterprise AI infrastructure, combining purpose-built models with the operational layer required to deploy, govern, observe, and scale AI across regulated environments.
Bigger models are not always better models. In production enterprise AI, the goal is not maximum model size. The goal is maximum accuracy per token, per dollar, and per workflow.
About AISquared
AISquared helps large organizations bring AI to where work happens. We do this by leveraging our SaaS or on-prem platform, which combines data sources with advanced AI/ML functionality and embeds intelligent insights into business applications. Trusted by the largest financial institutions, the most complicated supply chain logistics companies, and the U.S. Department of War, AISquared enables seamless collaboration between data science teams and business users, driving faster, more informed decision-making.
Contact
5WPR
aisquared@5wpr.com
