Enterprise search and AI company Glean is making a pointed claim: businesses that rely on Anthropic's Claude API are paying roughly 80% more than they need to. The assertion, reported by The Information, comes from Glean's own internal analysis of customer usage patterns and positions the company as a more cost-efficient alternative for enterprises building on top of large language models.
What Glean Is Claiming
According to the report, Glean argues that many enterprise customers send far more data to Claude than their applications actually require. Redundant context, unoptimized prompts, and inefficient call structures all inflate token counts, which directly drives up API costs. Glean says its platform handles prompt engineering and context management in ways that dramatically reduce those overheads. The 80% figure is striking, and Glean has a commercial interest in making the comparison, but the underlying issue of token inefficiency is one that developers across the industry have flagged for some time. This is not an isolated critique of one vendor's pricing; it reflects a broader challenge in how enterprises adopt AI at scale.
Key Facts
- Glean claims enterprise Claude users overpay by up to 80% on API bills
- The inefficiency stems from bloated prompts, excess context, and poor call optimization
- Glean positions its platform as a managed layer that reduces token waste
- The Information first reported the claim, citing Glean's internal analysis
- Anthropic has not publicly responded to the specific figures
The timing matters. AI infrastructure costs have become a flashpoint for CFOs approving technology budgets. Stories about runaway API bills are now common enough that some startups have built entire businesses around the problem. One earlier case covered here showed how an AI startup cut $30,000 in monthly bills by finding gaps in how providers structure their pricing. Glean's argument fits a pattern: the sticker price of a model is rarely the full story once enterprise usage scales up.
The way most enterprises use these APIs today, they are essentially paying for a lot of tokens that do nothing useful for the end user.Glean, via The Information
What This Means for Anthropic
Anthropic has been expanding its enterprise footprint aggressively. Claude's model family now covers a wide range of capability tiers, from lightweight options suited for high-volume tasks to the more powerful Sonnet and Opus variants used for complex reasoning work. That breadth is partly designed to give developers cost flexibility, but Glean's claim suggests the tooling and guidance around how to use those models efficiently may not be keeping pace with adoption.
It is worth noting that Glean competes directly with Anthropic for enterprise AI wallet share. The company is not a neutral auditor. Even so, the substance of the critique, that most enterprise teams lack the prompt engineering sophistication to use these APIs cost-effectively, is widely accepted among AI developers. The question is whether the gap is Anthropic's problem to solve or the customer's.
For enterprises currently evaluating or renegotiating their AI contracts, the practical takeaway is straightforward: token usage audits are worth doing before assuming your current spend is optimized. Managed platforms like Glean's offer one path, but engineering teams can also address inefficiency internally through prompt compression, caching, and smarter context windowing. The 80% figure may be Glean's best-case marketing number, but even a fraction of that savings would be material for companies running Claude at scale across thousands of daily users.
Anthropic has not issued a public response to the specific claims. Whether the company addresses the efficiency question directly, through documentation, tooling, or pricing restructuring, could shape how enterprise buyers view the platform heading into the second half of the year.
