Skip to content

Codebase intelligence for AI and humans: code health scores, auto-generated docs, git analytics, dead code detection, and architectural decisions via MCP.

Fix/minor (#60)

Swati Ahujac239c1544108
diff --git a/README.md b/README.md
index d8116ab04..bf1b62916 100644
--- a/README.md
+++ b/README.md
@@ -29,6 +29,23 @@ The result: Claude Code answers *"why does auth work this way?"* instead of *"he
---
+## πŸ† Benchmarked against frontier LLMs
+
+> **[repowise-bench β†’](https://github.com/repowise-dev/repowise-bench)** β€” an open SWE-QA benchmark that grades how well standard LLMs answer real software-engineering questions over real repositories.
+>
+> On 48 paired tasks from `pallets/flask` (`claude-sonnet-4-6`, end-to-end), repowise-augmented Claude Code matches baseline answer quality while being dramatically leaner:
+>
+> | Metric (per task, mean) | Baseline | **+ repowise** | Ξ” |
+> |---|---|---|---|
+> | πŸ’° Cost | $0.1396 | **$0.0890** | **βˆ’36 %** |
+> | ⚑ Wall time | 41.7 s | **33.9 s** | **βˆ’19 %** |
+> | πŸ› οΈ Tool calls | 7.4 | **3.8** | **βˆ’49 %** |
+> | πŸ“„ Files read | 1.9 | **0.2** | **βˆ’89 %** |
+>
+> **32 / 48 (67 %)** tasks are cheaper with repowise β€” at parity quality (judge Ξ” β‰ˆ βˆ’0.01).
+
+---
+
## What's new
### Faster indexing