{"slug": "gleans-ai-assistant-uses-70-fewer-tokens-than-anthropics-claude-cowork", "title": "Glean’s AI assistant uses 70% fewer tokens than Anthropic’s Claude Cowork", "summary": "Enterprise AI startup Glean announced on August 26, 2026, that its AI Assistant consumed 70% fewer tokens than Anthropic's Claude Cowork across more than 180 enterprise tasks, translating to an 81% reduction in per-task token costs, with Glean averaging $0.58 per task versus Claude Cowork's $2.98. Glean, which reached $300 million in Annual Recurring Revenue by May 2026 and closed a Series F round at $7.2 billion in June 2025, attributes the efficiency to its pre-indexed, permission-aware context graph and model routing across more than 40 models. The company also introduced the Tau desktop AI workspace alongside the benchmark announcement.", "body_md": "Photo: Steve A Johnson / Pexels\n\n# Glean’s AI assistant uses 70% fewer tokens than Anthropic’s Claude Cowork\n\nEnterprise AI startup claims 81% cost savings per task and growing customer preference as the race to cut corporate AI bills heats up\n\nEnterprise AI spending is becoming its own line-item crisis for corporate finance teams, and Glean just showed up with a calculator. The company announced on August 26, 2026, that its AI Assistant consumed 70% fewer tokens than Anthropic’s Claude Cowork across more than 180 enterprise tasks, translating that efficiency gap into an 81% reduction in per-task token costs.\n\nThe numbers are specific enough to take seriously. Glean’s platform averaged $0.58 per task in token costs. Claude Cowork averaged $2.98 for the same work.\n\n## How Glean pulls off the efficiency\n\nThe core mechanism behind Glean’s token savings is what the company calls a pre-indexed, permission-aware context graph. Because Glean pre-indexes enterprise data and tracks permissions in real time, it can retrieve relevant context without repeatedly querying expensive models from scratch.\n\nThat pre-indexed structure connects more than 275 business applications and maintains a live knowledge layer across them.\n\nThe second lever is model routing. Rather than defaulting every query to a premium frontier model, Glean automatically routes requests across more than 40 different models, selecting cheaper alternatives when the task doesn’t require heavy lifting.\n\nGlean also compared its remote setup against off-the-shelf Model Context Protocol tools used in Claude Cowork in earlier tests run between May and July 2026. Evaluators preferred Glean’s remote configuration roughly 2.5 times more than those baseline MCP tools.\n\nHuman graders reviewing outputs preferred Glean’s results 78% of the time when scoring for correctness, response quality, and overall interaction. Glean commissioned that evaluation, so some skepticism about methodology is fair, and independent reviewers did flag certain discrepancies in how the benchmarks were constructed. Still, the independent assessments did not dispute the core efficiency claim.\n\n## The business case behind the benchmark\n\nGlean reached $300 million in Annual Recurring Revenue by May 2026, up from just over $100 million roughly 15 months earlier.\n\nA Series F funding round closed in June 2025 placed Glean at $7.2 billion.\n\nThe new Tau desktop AI workspace, introduced alongside the benchmark announcement, provides enhanced model access and tighter usage controls at the desktop level, giving IT administrators more granularity over how AI tools get deployed and what they can access.\n\n## Who Glean is actually competing against\n\nThe benchmark comparison with Claude Cowork is a deliberate competitive move, but Glean’s real market battle is against Microsoft and Google, both of which have deeply integrated AI assistants into productivity suites that most enterprises already pay for. Microsoft Copilot sits inside Teams, Outlook, and Word. Google’s Gemini is woven into Workspace.\n\nGlean’s permission-aware architecture addresses a concern that matters to enterprise buyers: data control. When an AI assistant can see everything across 275 connected applications, the question of who controls what access becomes genuinely important. Glean’s real-time permission layer means employees only surface data they’re already authorized to see.\n\nThe $0.58 versus $2.98 per-task comparison will land differently for a company running 50,000 AI queries per day than for one running 500, and Glean’s growth numbers suggest it’s already reaching organizations at that scale.\n\n**Disclosure:** This article was edited by Editorial Team. For more information on how we create and review content, see our\n\n[Editorial Policy](https://cryptobriefing.com/editorial-policy/).", "url": "https://wpnews.pro/news/gleans-ai-assistant-uses-70-fewer-tokens-than-anthropics-claude-cowork", "canonical_source": "https://cryptobriefing.com/glean-ai-assistant-70-percent-fewer-tokens-claude/", "published_at": "2026-09-02 14:37:34+00:00", "updated_at": "2026-09-02 14:56:48.989396+00:00", "lang": "en", "topics": ["artificial-intelligence", "ai-products", "ai-infrastructure", "ai-tools"], "entities": ["Glean", "Anthropic", "Claude Cowork", "Microsoft", "Google", "Tau"], "alternates": {"html": "https://wpnews.pro/news/gleans-ai-assistant-uses-70-fewer-tokens-than-anthropics-claude-cowork", "markdown": "https://wpnews.pro/news/gleans-ai-assistant-uses-70-fewer-tokens-than-anthropics-claude-cowork.md", "text": "https://wpnews.pro/news/gleans-ai-assistant-uses-70-fewer-tokens-than-anthropics-claude-cowork.txt", "jsonld": "https://wpnews.pro/news/gleans-ai-assistant-uses-70-fewer-tokens-than-anthropics-claude-cowork.jsonld"}}