Previous edition: Prompt-cache hit rate by coding agent, April 2026
Cache hit rate by coding agent, October 2026
Cache hit rate by coding agent, October 2026
cached_tokens / input_tokens. Higher cache hit = lower effective input cost.
In October 2026 Cline led at 96% of input tokens from cache.
Which coding agents get the most from prompt caching? In October 2026 Cline led at 96% of input tokens from cache. Qwen Code was lowest at 82%. Most agents sat above 90%.
Why it mattersCoding agents resend large contexts on every turn, so cache hit rate drives their cost. A few points of cache hit rate is a large share of the input bill.
Period
Oct 2026
Updated
November 11, 2026
ID
coding-agent-cache-oct26
§ 01
Key findings
- 01Cline: 96%.
- 02Roo Code: 96%.
- 03GitHub Copilot: 95%.
- 04Cursor: 95%.
- 05Pi: 95%.
§ 01b
Questions this answers
- Which coding agent has the best cache hit rate?
- How much do coding agents rely on prompt caching?
- Why is Claude Code cache hit rate lower than other agents?
§ 02
Data
| Agent | Cache hit rate(percent) |
|---|---|
| Cline | 96.00% |
| Roo Code | 95.57% |
| GitHub Copilot | 95.45% |
| Cursor | 95.42% |
| Pi | 94.58% |
| Kilo Code | 93.82% |
| OpenCode | 93.75% |
| Hermes Agent | 92.78% |
| Zed | 91.98% |
| Zoo Code | 89.66% |
| Claude Code | 85.22% |
| Qwen Code | 81.87% |
§ 03
Cite as
APA
Click to copy
BibTeX
Click to copy
ID: coding-agent-cache-oct26·Updated November 11, 2026·Period Oct 2026