| LLM,Retriever,Submitted By,Scaffold,Accuracy (%),Recall (%),Search Calls,Calibration Error (%),Model Size (B),Open Weights?,LLM Link,Retriever Link,Evaluation Date,Submittor Link,Accuracy (GPT-4.1 Judge) |
| GPT-4.1,BM25,BrowseComp-Plus,,15.30,16.42,10.35,68.52,-,No,https://openai.com/index/gpt-4-1/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,14.58 |
| GPT-4.1,Qwen3-Embed-8B,BrowseComp-Plus,,36.39,36.89,8.67,53.79,-,No,https://openai.com/index/gpt-4-1/,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,35.42 |
| o3,BM25,BrowseComp-Plus,,50.48,56.64,25.93,14.48,-,No,https://openai.com/index/introducing-o3-and-o4-mini/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,49.28 |
| o3,Qwen3-Embed-8B,BrowseComp-Plus,,65.90,73.24,23.97,19.68,-,No,https://openai.com/index/introducing-o3-and-o4-mini/,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,63.49 |
| Sonnet 4,BM25,BrowseComp-Plus,,14.70,21.31,9.95,28.78,-,No,https://www.anthropic.com/news/claude-4,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,14.34 |
| Sonnet 4,Qwen3-Embed-8B,BrowseComp-Plus,,37.35,47.33,9.03,24.84,-,No,https://www.anthropic.com/news/claude-4,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,36.75 |
| Opus 4,BM25,BrowseComp-Plus,,15.54,22.96,11.22,22.30,-,No,https://www.anthropic.com/news/claude-4,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,15.54 |
| Opus 4,Qwen3-Embed-8B,BrowseComp-Plus,,36.75,50.84,10.24,12.33,-,No,https://www.anthropic.com/news/claude-4,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,36.14 |
| Gemini 2.5 Flash,BM25,BrowseComp-Plus,,16.27,21.45,10.56,28.51,-,No,https://arxiv.org/abs/2507.06261,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,15.54 |
| Gemini 2.5 Flash,Qwen3-Embed-8B,BrowseComp-Plus,,34.58,40.19,9.77,20.42,-,No,https://arxiv.org/abs/2507.06261,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,33.01 |
| Gemini 2.5 Pro,BM25,BrowseComp-Plus,,19.88,22.81,7.44,52.45,-,No,https://arxiv.org/abs/2507.06261,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,19.04 |
| Gemini 2.5 Pro,Qwen3-Embed-8B,BrowseComp-Plus,,29.52,35.31,6.04,44.25,-,No,https://arxiv.org/abs/2507.06261,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,28.67 |
| Qwen3-32B,BM25,BrowseComp-Plus,,3.61,3.12,0.92,58.17,32,Yes,https://arxiv.org/abs/2505.09388,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,3.49 |
| Qwen3-32B,Qwen3-Embed-0.6B,BrowseComp-Plus,,4.22,3.45,0.91,61.20,32,Yes,https://arxiv.org/abs/2505.09388,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,4.10 |
| Qwen3-32B,Qwen3-Embed-4B,BrowseComp-Plus,,8.07,6.20,0.89,61.17,32,Yes,https://arxiv.org/abs/2505.09388,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,7.83 |
| Qwen3-32B,Qwen3-Embed-8B,BrowseComp-Plus,,10.72,7.80,0.94,58.88,32,Yes,https://arxiv.org/abs/2505.09388,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,10.36 |
| Qwen3-32B,ReasonIR,BrowseComp-Plus,,9.28,7.59,0.91,56.08,32,Yes,https://arxiv.org/abs/2505.09388,https://arxiv.org/abs/2504.20595,"Aug 8, 2025",,9.16 |
| SearchR1-32B,BM25,BrowseComp-Plus,,4.11,2.61,1.78,-,32,Yes,https://arxiv.org/abs/2505.15117,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,3.86 |
| SearchR1-32B,Qwen3-Embed-0.6B,BrowseComp-Plus,,6.02,5.30,1.73,-,32,Yes,https://arxiv.org/abs/2505.15117,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,5.66 |
| SearchR1-32B,Qwen3-Embed-4B,BrowseComp-Plus,,9.28,7.90,1.68,-,32,Yes,https://arxiv.org/abs/2505.15117,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,9.40 |
| SearchR1-32B,Qwen3-Embed-8B,BrowseComp-Plus,,11.08,10.17,1.69,-,32,Yes,https://arxiv.org/abs/2505.15117,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,10.36 |
| SearchR1-32B,ReasonIR,BrowseComp-Plus,,9.31,8.37,1.74,-,32,Yes,https://arxiv.org/abs/2505.15117,https://arxiv.org/abs/2504.20595,"Aug 8, 2025",,9.43 |
| oss-20b-low,BM25,BrowseComp-Plus,,3.99,5.36,1.89,41.56,20,Yes,https://openai.com/index/introducing-gpt-oss/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,4.11 |
| oss-20b-low,Qwen3-Embed-8B,BrowseComp-Plus,,14.10,17.37,1.87,36.17,20,Yes,https://openai.com/index/introducing-gpt-oss/,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,13.37 |
| oss-20b-medium,BM25,BrowseComp-Plus,,16.87,21.96,13.72,41.19,20,Yes,https://openai.com/index/introducing-gpt-oss/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,16.39 |
| oss-20b-medium,Qwen3-Embed-8B,BrowseComp-Plus,,30.48,41.31,13.64,35.49,20,Yes,https://openai.com/index/introducing-gpt-oss/,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,29.88 |
| oss-20b-high,BM25,BrowseComp-Plus,,21.45,31.98,26.87,33.08,20,Yes,https://openai.com/index/introducing-gpt-oss/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,21.08 |
| oss-20b-high,Qwen3-Embed-8B,BrowseComp-Plus,,35.06,49.29,23.87,27.69,20,Yes,https://openai.com/index/introducing-gpt-oss/,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,34.58 |
| oss-120b-low,BM25,BrowseComp-Plus,,9.76,8.54,2.06,43.70,120,Yes,https://openai.com/index/introducing-gpt-oss/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,9.52 |
| oss-120b-low,Qwen3-Embed-8B,BrowseComp-Plus,,25.54,22.50,2.21,40.37,120,Yes,https://openai.com/index/introducing-gpt-oss/,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,24.94 |
| oss-120b-medium,BM25,BrowseComp-Plus,,24.58,27.02,9.73,45.70,120,Yes,https://openai.com/index/introducing-gpt-oss/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,23.73 |
| oss-120b-medium,Qwen3-Embed-8B,BrowseComp-Plus,,38.55,43.45,9.64,41.63,120,Yes,https://openai.com/index/introducing-gpt-oss/,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,37.59 |
| oss-120b-high,BM25,BrowseComp-Plus,,29.16,35.50,19.45,45.92,120,Yes,https://openai.com/index/introducing-gpt-oss/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,28.67 |
| oss-120b-high,Qwen3-Embed-8B,BrowseComp-Plus,,44.10,52.63,18.35,39.32,120,Yes,https://openai.com/index/introducing-gpt-oss/,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,42.89 |
| GPT-5,BM25,BrowseComp-Plus,,57.59,61.70,23.23,12.63,-,No,https://openai.com/index/introducing-gpt-5/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,55.90 |
| GPT-5,Qwen3-Embed-8B,BrowseComp-Plus,,71.69,78.98,21.74,9.58,-,No,https://openai.com/index/introducing-gpt-5/,https://arxiv.org/abs/2506.05176,"Aug 8, 2025",,70.12 |
| kimi-k2-0711-preview,BM25,BrowseComp-Plus,,14.58,15.43,14.90,69.41,1000,Yes,https://moonshotai.github.io/Kimi-K2/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 16, 2025",,13.49 |
| kimi-k2-0711-preview,Qwen3-Embed-8B,BrowseComp-Plus,,35.42,38.38,11.22,53.51,1000,Yes,https://moonshotai.github.io/Kimi-K2/,https://arxiv.org/abs/2506.05176,"Aug 16, 2025",,34.22 |
| DeepSeek-R1-0528,BM25,BrowseComp-Plus,,8.56,7.58,2.71,62.47,685,Yes,https://api-docs.deepseek.com/news/news250528,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 16, 2025",,8.44 |
| DeepSeek-R1-0528,Qwen3-Embed-8B,BrowseComp-Plus,,16.39,16.32,2.72,57.66,685,Yes,https://api-docs.deepseek.com/news/news250528,https://arxiv.org/abs/2506.05176,"Aug 16, 2025",,16.02 |
| WebSailor-32B,BM25,BrowseComp-Plus,,12.05,12.00,8.76,-,32,Yes,https://arxiv.org/pdf/2507.02592,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Sep 28, 2025",,11.20 |
| WebSailor-32B,Qwen3-Embed-8B,BrowseComp-Plus,,27.59,29.30,8.00,-,32,Yes,https://arxiv.org/pdf/2507.02592,https://arxiv.org/abs/2506.05176,"Sep 28, 2025",,25.78 |
| Tongyi-DeepResearch-30B-A3B,BM25,BrowseComp-Plus,,33.98,46.83,32.92,-,30,Yes,https://tongyi-agent.github.io/blog/introducing-tongyi-deep-research/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Sep 28, 2025",,33.37 |
| Tongyi-DeepResearch-30B-A3B,Qwen3-Embed-8B,BrowseComp-Plus,,44.46,62.32,30.37,-,30,Yes,https://tongyi-agent.github.io/blog/introducing-tongyi-deep-research/,https://arxiv.org/abs/2506.05176,"Sep 28, 2025",,44.10 |
| FrugalRAG-Qwen7B,Qwen3-Embed-8B,FrugalRAG,,20.46,23.57,7.95,73.67,7,No,https://arxiv.org/abs/2507.07634,https://arxiv.org/abs/2506.05176,"Sep 22, 2025",https://arxiv.org/abs/2507.07634,- |
| GLM-4.6,BM25,BrowseComp-Plus,,27.74,34.98,29.71,37.12,357,Yes,https://docs.z.ai/guides/llm/glm-4.6,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Oct 13, 2025",,- |
| GLM-4.6,Qwen3-Embed-8B,BrowseComp-Plus,,48.19,57.72,25.6,26.3,357,Yes,https://docs.z.ai/guides/llm/glm-4.6,https://arxiv.org/abs/2506.05176,"Oct 13, 2025",,- |
| Gemini 2.5 Flash,Mixedbread Search,Mixedbread,,39.88,42.78,8.3,24.61,-,No,https://openai.com/index/introducing-gpt-5/,https://www.mixedbread.com/,"Oct 22, 2025",https://www.mixedbread.com/,38.19 |
| GPT-5,Mixedbread Search,Mixedbread,,74.1,80.29,21.22,11,-,No,https://openai.com/index/introducing-gpt-5/,https://www.mixedbread.com/,"Oct 22, 2025",https://www.mixedbread.com/,73.25 |
| oss-120b-high,BM25,BrowseComp-Plus,,29.16,35.50,19.45,45.92,120,Yes,https://openai.com/index/introducing-gpt-oss/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Aug 8, 2025",,28.67 |
| GPT-5,Mixedbread Search,Agentica,"Agentica custom scaffold: Allows agents to execute arbitrary code in a REPL. Agents had access to `search` and `get_doc_by_id` (avg. 8.83 calls this run) functions. `search` returned top 200 results instead of 5 in Standard scaffold. The submitter calculated recall using docs that the agent fetched using `get_doc_by_id`, due to the high top k used in `search`.",78.41,48.85,44.67,7.92,-,No,https://openai.com/index/introducing-gpt-5/,https://www.mixedbread.com/,"Dec 11, 2025",https://www.symbolica.ai/agentica,77.11 |
| oss-120b-high,BM25,BrowseComp-Plus,"gpt-oss native toolset: `search`, `open` (view full doc by id, avg. 11.17 calls this run), `find` (grep in doc, avg. 4.23 calls this run)",43.61,58.38,46.51,36.82,120,Yes,https://openai.com/index/introducing-gpt-oss/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Dec 31, 2025",, |
| oss-120b-high,Qwen3-Embed-8B,BrowseComp-Plus,"gpt-oss native toolset: `search`, `open` (view full doc by id, avg. 11.06 calls this run), `find` (grep in doc, avg. 4.28 calls this run)",56.99,74.7,41.86,29.05,120,Yes,https://openai.com/index/introducing-gpt-oss/,https://arxiv.org/abs/2506.05176,"Dec 31, 2025",, |
| GLM-4.7,BM25,BrowseComp-Plus,,33.25,45.97,41.55,16.9,358,Yes,https://z.ai/blog/glm-4.7,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Dec 26, 2025",, |
| GLM-4.7,Qwen3-Embed-8B,BrowseComp-Plus,,50.18,68.69,35.32,13.2,358,Yes,https://z.ai/blog/glm-4.7,https://arxiv.org/abs/2506.05176,"Dec 26, 2025",, |
| Tongyi-DeepResearch-30B-A3B,BM25,BrowseComp-Plus,"Tongyi-DeepResearch native toolset: `search`, `visit` (view full doc by id, avg. 2.75 calls this run)",36.87,43.02,17.91,-,30,Yes,https://tongyi-agent.github.io/blog/introducing-tongyi-deep-research/,http://trec.nist.gov/pubs/trec3/papers/city.ps.gz,"Dec 27, 2025",, |
| Tongyi-DeepResearch-30B-A3B,Qwen3-Embed-8B,BrowseComp-Plus,"Tongyi-DeepResearch native toolset: `search`, `visit` (view full doc by id, avg. 3.11 calls this run)",51.93,60.51,17.14,-,30,Yes,https://tongyi-agent.github.io/blog/introducing-tongyi-deep-research/,https://arxiv.org/abs/2506.05176,"Dec 27, 2025",, |
| GPT-5-mini,openJiuwen-deepsearch,openJiuwen,"openJiuwen custom setup: Multi-agent scaffold. Retrieved results are chunked; only most relevant chunks are returned in search, instead of truncating the full document by 512 tokens",80,54.51,6.05,18.82,-,No,https://openai.com/index/introducing-gpt-5/,https://openjiuwen.com/,"Nov 26, 2025",https://openjiuwen.com/en/, |
| OpenResearcher,Qwen3-Embed-8B,OpenResearcher,"OpenResearcher toolset: `search` (with lucene highlighter to filter raw contents), `open` (view full doc by id, avg. 14.46 calls this run), `find` (grep in doc, avg. 5.04 calls this run)",60.92,75.77,41,27.59,30,Yes,https://github.com/TIGER-AI-Lab/OpenResearcher,https://arxiv.org/abs/2506.05176,"Feb 12, 2026",https://github.com/TIGER-AI-Lab/OpenResearcher, |
| SearchAgent-Zero,Qwen3-Embed-8B,SearchAgent-Zero,,37.95,50.87,38.47,58.06,8,Yes,https://github.com/NLPJCL/SearchAgent-Zero,https://arxiv.org/abs/2506.05176,"Feb 24, 2026",https://github.com/NLPJCL/SearchAgent-Zero, |
| Afficher plus17 marsStructured experimental results into CSV format with consistent field mappingStructured experimental results into CSV format with consistent field mappingHere are the CSV rows. I've used the same LLM links as existing entries, left the new retriever links blank (fill in when available), and used the scaffold column to distinguish "with get" variants: |
| oss-120b-high,GTE-ModernColBERT,LightOn,,55.66,66.94,17.51,31.02,120,Yes,https://openai.com/index/introducing-gpt-oss/,https://huggingface.co/lightonai/GTE-ModernColBERT-v1,"March 16, 2026",https://lighton.ai/, |
| oss-120b-high,GTE-ModernColBERT,LightOn,Standard scaffold + get_document (avg. 1.76 calls),58.92,57.99,13.29,32.84,120,Yes,https://openai.com/index/introducing-gpt-oss/,https://huggingface.co/lightonai/GTE-ModernColBERT-v1,"March 16, 2026",https://lighton.ai/, |
| oss-120b-high,Reason-ModernColBERT,LightOn,,59.04,68.64,18.87,,120,Yes,https://openai.com/index/introducing-gpt-oss/,https://huggingface.co/lightonai/Reason-ModernColBERT,"March 16, 2026",https://lighton.ai/, |
| oss-120b-high,Reason-ModernColBERT,LightOn,Standard scaffold + get_document (avg. 1.96 calls),61.20,60.84,13.87,32.45,120,Yes,https://openai.com/index/introducing-gpt-oss/,https://huggingface.co/lightonai/Reason-ModernColBERT,"March 16, 2026",https://lighton.ai/, |
| GPT-5,Reason-ModernColBERT,LightOn,,79.52,83.52,19.31,7.46,-,No,https://openai.com/index/introducing-gpt-5/,https://huggingface.co/lightonai/Reason-ModernColBERT,"March 16, 2026",https://lighton.ai/, |
| GPT-5,Reason-ModernColBERT,LightOn,Standard scaffold + get_document (avg. 3.00 calls),87.59,81.55,13.27,6.07,-,No,https://openai.com/index/introducing-gpt-5/,https://huggingface.co/lightonai/Reason-ModernColBERT,"March 16, 2026",https://lighton.ai/, |
|
|
