AI search startup Glean is entering the white-collar office automation space, competing directly with players like Anthropic, and its executives are zeroing in on two issues that keep corporate chief information officers up at night: token costs and data security. Last week, Glean published internal benchmark results showing that its own AI assistant is significantly cheaper to operate than Anthropic's Claude Cowork application when executing over 180 business tasks.
Glean attributes this cost advantage primarily to a so-called context layer it builds for clients such as Intuit, Booking.com, and Zillow, which allows AI agents to rapidly retrieve the data needed to complete tasks. This layer includes an "enterprise knowledge graph" that helps the AI understand relationships between various types of internal corporate data.
A Glean spokesperson provided a concrete example: when using Anthropic's Claude Opus 4.8 model as the underlying engine, Glean's AI assistant consumes 70% fewer tokens to complete identical tasks compared to Anthropic's native Claude Cowork application, which runs on the high-reasoning Claude Sonnet 5 model. Notably, Anthropic's own website describes Sonnet 5 as having performance "close to Opus 4.8" but at a lower cost point.
Overall, Glean claims that on a per-task billing basis, its AI assistant charges clients an average of 81% less than Claude Cowork. This is partly thanks to Glean's built-in scheduling software, which automatically routes tasks to the most appropriate model based on complexity. An Anthropic spokesperson declined to comment on these figures.
Glean completed its last funding round in June of last year at a post-money valuation of $7.2 billion. CEO Arvind Jain explained that the context layer eliminates lengthy retrieval processes AI would otherwise perform to locate target data, thereby slashing expenses for customers. "Last year, AI costs weren't particularly high because the tasks AI handled were limited in complexity. But now AI is tackling a large volume of complex work, runtime has stretched longer, and costs have naturally risen," he said.
In May, the company announced that its annual recurring revenue (contract revenue over the next 12 months) had already surpassed $300 million. Glean's sharp focus on AI usage costs also highlights how this seven-year-old startup plans to carve out a niche in the fiercely competitive enterprise AI agent market, where rivals include major cloud providers as well as Anthropic and OpenAI.
Meanwhile, Glean states that when its agents operate in enterprise environments, they can achieve the same level of security and data protection standards that companies extend to their internal employees. In June, Glean launched several virtual colleague-style agents, including a duty operations engineer agent deployed on Slack that is designed to keep enterprise applications running smoothly. If an application fails, this agent gathers fault details, diagnoses the root cause, and outlines viable remediation plans.
Jain also outlined several steps Glean is taking to enhance the enterprise safety of its agents: assigning each AI agent a distinct identity and authentication credentials, strictly limiting it to access only authorized data. "These agents run for extended periods and have the permissions and capabilities to execute various operations within systems. Without proper governance, they can easily spiral into uncontrolled actions," Jain noted.