Quotient Labs

Fermat's Last Token is a plugin from Quotient Labs, a company that states it is backed by Y Combinator, that reduces token spend when using Claude Code. It runs alongside Claude Code without changing the models or the way a developer works, and is installed with a single command. The vendor describes three techniques. Semantic context pruning optimizes older context while keeping recent instructions, tool relationships and task critical state, and it works with Anthropic's prompt caching. Optimized agent tools do more per call, reduce repeated file reads and tie edits to files the agent has actually inspected. Quality preserving compression shrinks tool outputs and earlier assistant prose before cache writes, and passes content through unchanged when compression would be unsafe. The site reports its own benchmark, called SessionBench, built on refactoring tickets from Scale AI's SWE Atlas. Across 15 paired sessions in three repositories, run with Claude Opus 5, it reports average agent cost falling from $51.67 to $27.29, a 47 percent saving, with average quality scores rising from 0.83 to 0.92. By repository the reported savings range from 34 percent to 51 percent. The methodology, raw logs and a reproduction repository are linked from the page. The audience is developers and teams paying for Claude Code usage. Plan prices are not published on the site, which points users to a dashboard and a free trial.