SuperMaven: 300k-token autocomplete engine, 10x faster than Copilot
In one sentence Jacob Jackson, Tabnine co-founder, launches SuperMaven: a code autocomplete engine with 300k-token context window, 10x lower latency than Copilot, treating completion as a long-context retrieval problem. Later acquired by Cursor.
One of the most frustrating problems with code completion tools like GitHub Copilot is that when you work on long files, or when you have many related files open in your editor, the tool "forgets" what you wrote earlier or in other files. It is like an assistant with a very short memory.
Jacob Jackson, who had already co-founded Tabnine, tackled this problem differently with SuperMaven. Instead of using a large generic model, he built an engine specialized only for autocomplete with an enormously larger context window: 300,000 tokens, roughly 15 times larger than the competition at the time.
The practical result: SuperMaven could see the entire contents of long files and all related files open in the editor at the same time. Not only was suggestion quality better on large codebases, but response speed was about 10 times faster than Copilot thanks to the specialized architecture. Cursor, the popular AI editor, later acquired SuperMaven, integrating this technology into its autocomplete engine.
Companies
SuperMaven
Tools
—
Tags
Sources