ByteAsk 的 C/C++ 编码智能体通过上下文引擎在模型读取前定位修改位置，将每次运行的 token 从 1,795K 降至 395K，减少 78%，修复 bug 数量不变且成本降低 68.1%。在冷门项目和难题上节省约 80%。该工具此前已发布，主打多 SWE-bench（C/C++）SOTA、比 cc/codex 少 60% token，提供免费方案并开源。
Another great launch to reduce token cost.
Most of what you pay an AI agent to fix C++ goes to finding the code, before any fixing.
@ByteAsk removed 78% of the tokens and fixed exactly as many bugs and at 68.1% lower cost.
ByteAsk is a C and C++ coding agent that finds where a fix belongs before the AI model starts reading. Before the model starts, ByteAsk’s context engine reads the code as the build sees it and marks where the change belongs.
It also flags sibling sites needing the same edit, and the agent checks its patch against only the tests covering them.
That upfront map cut pooled tokens from 1,795K to 395K per run, with about 80% savings on little-known projects and hard issues.
在 X 查看被引用的帖子Introducing @ByteAsk
Your AI C++ engineer, and a swarm of them
SOTA on multi SWE bench (C/C++)
60% fewer tokens wrt cc/codex
auto debug/perf/optimize/test with native C/C++ toolsFree plan. Open Source.
github.com/ByteAsk/cli
来源：Rohan Paul · x.com