Skip to content

~95% on SimpleQA (e.g. Qwen3.6-27B on a 3090). Supports all local and cloud LLMs (llama.cpp, Ollama, Google, ...). 10+ search engines - arXiv, PubMed, your private documents. Everything Local & Encrypted.

docs(readme): replace stale benchmark table with current local-LLM results (#3991)

LearningCircuitb066007ec70b
diff --git a/README.md b/README.md
index ab65b6109d..7c6fe1f73f 100644
--- a/README.md
+++ b/README.md
@@ -422,15 +422,16 @@ search(query="copyright fair use precedents", engine="openclaw")
## 📊 Performance & Analytics
### Benchmark Results
-Early experiments on small SimpleQA dataset samples:
-| Configuration | Accuracy | Notes |
-|--------------|----------|--------|
-| gpt-4.1-mini + SearXNG + focused_iteration | 90-95% | Limited sample size |
-| gpt-4.1-mini + Tavily + focused_iteration | 90-95% | Limited sample size |
-| gemini-2.0-flash-001 + SearXNG | 82% | Single test run |
+Headline results from the [community benchmarks](https://huggingface.co/datasets/local-deep-research/ldr-benchmarks) using the `langgraph-agent` strategy with Serper search, fully local via Ollama:
-Note: These are preliminary results from initial testing. Performance varies significantly based on query types, model versions, and configurations. [Run your own benchmarks →](docs/BENCHMARKING.md)
+| Model | SimpleQA | xbench-DeepSearch |
+|---|---|---|
+| Qwen3.6-27B | 95.7% (287/300) | 77.0% (77/100) |
+| Qwen3.5-9B | 91.2% (182/200) | 59.0% (59/100) |
+| gpt-oss-20B | 85.4% (295/346) | – |
+
+Caveats: small samples, LLM-grader noise, and SimpleQA contamination risk on newer base models. See the [HF dataset card](https://huggingface.co/datasets/local-deep-research/ldr-benchmarks) for the full leaderboard, or [run your own benchmarks →](docs/BENCHMARKING.md).
**Full community leaderboard:** The community maintains a growing collection of benchmark results across models, strategies, and search engines in a dedicated repo with CI-validated submissions and auto-generated leaderboards: