ByteAsk Paper: Build-Aware C++ Agent Cuts Tokens 63-80%, Matches Codex CLI Fixes

rohanpaul_ai · x · 2026-11-11

ByteAsk Research published a paper on ByteAsk, a coding agent built for C and C++. Its core is a build-aware context engine: it understands the codebase the way the project's own build does, telling the model where a change belongs, which related sites need the same change, and which tests cover them — then checks the change against only those tests.\n\nCompared with OpenAI's Codex CLI running the same underlying model on recent issues from popular and little-known C++ projects plus a hard set:\n\n- Both agents fixed the same number of runs (129 of 144 each across the two main sets);\n- ByteAsk used 63-80% fewer tokens and cost 59-70% less on every task set;\n- Savings peak at 80% fewer tokens on little-known, hard C++ projects.\n\nTakeaway: for build-dependent C++ codebases, making the agent understand the build before writing code slashes cost with zero loss in fix rate. Code and tasks are on GitHub.

Related event: ByteAsk Open-Sources C++ Coding Agent That Cuts Token Use by Up to 80%(2 posts)→

Original post →

More from coding & agent

coding & agent channel →