Previous edition: Tool-call finish rate by coding agent, April 2026
Tool-call finish rate by coding agent, October 2026
Tool-call finish rate by coding agent, October 2026
Share of API calls that end with a tool_calls finish reason (the primary agentic pattern).
tool_calls share
In October 2026 Roo Code ended 95% of calls with a tool call, the highest share.
How agentic is each coding agent? In October 2026 Roo Code ended 95% of calls with a tool call, the highest share. Cline was lowest at 45%.
Why it mattersA high tool_calls share means the agent runs long loops of edits and commands. A higher stop share means more single-turn answers.
Period
Oct 2026
Updated
November 11, 2026
ID
coding-agent-finish-oct26
§ 01
Key findings
- 01Roo Code: 95% tool_calls.
- 02Zoo Code: 95% tool_calls.
- 03GitHub Copilot: 89% tool_calls.
- 04Cursor: 89% tool_calls.
- 05Pi: 87% tool_calls.
§ 01b
Questions this answers
- Which coding agent makes the most tool calls?
- How often do coding agent calls end in a tool call?
- Which agents favor single-turn completions?
§ 02
Data
| Agent | tool_calls rate(percent) |
|---|---|
| Roo Code | 95.35% |
| Zoo Code | 95.08% |
| GitHub Copilot | 89.26% |
| Cursor | 88.80% |
| Pi | 86.84% |
| Zed | 85.61% |
| OpenCode | 85.08% |
| Kilo Code | 81.11% |
| Hermes Agent | 76.00% |
| Qwen Code | 73.14% |
| Claude Code | 60.96% |
| Cline | 45.11% |
§ 03
Cite as
APA
Click to copy
BibTeX
Click to copy
ID: coding-agent-finish-oct26·Updated November 11, 2026·Period Oct 2026