Glean launches Tau to cut enterprise AI token costs

Glean has launched Glean Tau, a desktop AI workspace for enterprise users, and released a new benchmark showing its assistant used fewer tokens and was preferred more often than Claude Cowork.

Glean Tau is designed to let users work across local files, applications and code while drawing on what Glean describes as enterprise context. The launch comes as businesses scrutinise AI spending more closely, particularly the costs tied to model usage and employee oversight.

The product extends Glean's existing workplace AI system from enterprise software and data sources to the desktop. It can carry out multi-step tasks including organising files, analysing documents, creating spreadsheets and working across connected applications.

Glean linked the launch to what it calls "botsitting", the time workers spend supervising and correcting AI systems. Research from its Work AI Institute, based on a survey of 6,000 full-time digital workers, found employees spend 6.4 hours a week on that work, more time than they spend using AI productively.

The benchmark results released with the product announcement focused on token costs during common workplace tasks. Glean said it tested more than 180 enterprise tasks and compared Glean Assistant, with automatic model routing enabled, against Claude Cowork using Claude Sonnet 5.

According to Glean, its assistant averaged USD $0.58 per task, compared with USD $2.98 for Claude Cowork, an 81% reduction in token costs.

Token use was also lower in the comparison. Glean said its system used 1.3 million tokens versus 4.4 million for Claude Cowork, a reduction of about 70%.

Glean attributed the difference to its use of a pre-indexed, permission-aware view of company data rather than repeatedly searching connected systems for each task. It also pointed to automatic model routing, which selects different models and reasoning settings depending on the work involved.

On output quality, Glean said graders preferred its answers 78% of the time across tasks spanning sales, engineering, marketing, human resources, product, finance and customer support. The assessment looked at overall preference, correctness, completeness and interaction quality.

The comparison reflects a broader shift in how large companies judge AI systems. Cost control, not just model quality, has become a more prominent issue as organisations deploy assistants and agents across more of their operations.

Recent disclosures by major technology users have added to that pressure. Uber and ServiceNow have both publicly discussed how quickly annual AI budgets were consumed, highlighting how token-intensive tools can affect spending assumptions.

Cost pressure

Glean's broader argument is that the economics of workplace AI depend heavily on context management, not model choice alone. In its view, systems without ready access to enterprise information consume more tokens because they must repeatedly search, retrieve and reason through fragmented data.

That issue has become more visible as companies experiment with multiple AI models at once. The mix now includes frontier systems from established suppliers and newer open models at lower price points, prompting businesses to compare cost and output more systematically.

Glean said it has also been analysing a wider set of enterprise tasks across dozens of models to guide its routing decisions. It argues that no single model is best for every use case and that balancing quality with price is becoming a core management task for IT teams.

Broader rollout

Alongside Glean Tau, the company outlined a broader set of additions to its workplace AI platform. These include tools for identifying where AI can take on work, interactive dashboards that combine structured and unstructured business data, shared team chat threads, expanded AI governance functions and systems intended to detect risky agent behaviour in context.

Some of those functions are generally available, while others remain in beta or are due later. Glean Tau itself is listed as coming soon.

Glean also framed governance as a central selling point for larger employers. It said the same context layer used to improve task completion can also be applied to permissions, policy controls and monitoring how AI systems act across internal data and tools.

That approach appears aimed at buyers trying to avoid a fragmented AI estate, where separate assistants, agents and model interfaces create duplicated costs and oversight burdens. As more workplace tools add AI features, many organisations are seeking a common control layer rather than managing each service in isolation.

Emrecan Dogan, Chief Product Officer at Glean, said the company's approach rests on that premise. "Models are getting better and more interchangeable. Understanding the enterprise is not. We designed Glean around that reality in 2019. The more AI moves from answering questions to actually doing work, the more valuable company context becomes, because it tells AI what matters, what should happen next, and how to act within the realities of the business," Dogan said.

Customer references supplied by Glean echoed that focus on control and data context. "Glean has made it easier for us to scale AI safely by giving teams stronger control, trusted context, and a multi-model choice, so they can move faster without giving up governance," said Rhonda Baldwin, CIO, LaunchDarkly.

A second customer described token efficiency as a differentiator. "Glean stands out because it delivers context above all else. In other platforms, it would take a huge number of tokens to recreate the same depth of context. Because Glean is native and the data is already indexed and loaded, the value is immediate," said Lee.