ColGrep-Agent-2B cuts Claude Code's token usage by 34.3% vs grep across 13 repos
IgorCarron · x · 2026-10-12
- Benchmark: Across 52 code-search questions on 13 repositories and 624 benchmark sessions, ColGrep-Agent-2B reduced Claude Code's API token usage compared to plain grep.
- Results: 34.3% mean token savings overall (up to 46.3% on single-location searches, 31.1% cross-folder); it used fewer tokens on 88% of questions, saving 49k tokens per question on average.
- Takeaway: Swapping naive grep-based search inside coding agents for a small retrieval model can meaningfully cut inference costs without changing the main model.
More from coding & agent
- SubQ Snap launches as a long-context decision model: 384K tokens, decisions in ~2-3s at $0.05/M — ryanallred · 2026-10-13
- One wrong 'merge it' prompt shipped ts-rust to T3 Code for 8 hours unnoticed — zeeg · 2026-10-13
- Don't specify languages or architecture for new projects — let the model research and choose — sytelus · 2026-10-13
- GPT-6 one-shots a custom IR blaster app to replace a dead AC remote — PrasoonPratham · 2026-10-13
- NVIDIA's NURBS-based CAD kernel SMLIB goes open source, with MCP agent support — burhop · 2026-10-13
- Why checking errors is bad: Jimmy Koppel's contrarian take on defensive coding — jimmykoppel · 2026-10-13