FACTS Search V2
We show this table for reference; we do not rank on it.
FACTS Search V2 measures factual answering with a shared Brave Search tool. Overall, public, private, hedging, search-count, attempted-accuracy, and F1 measurements remain distinct.
Evaluation results published by Google DeepMind, Google Research, and Kaggle. How Kaggle Benchmarks reruns tests.
Kaggle labels this benchmark Apache 2.0. We retain aggregate measurements with attribution and source links. That declaration does not establish rights to third-party documents, videos, or other underlying task material; no task contents are included. Original article terms.
Published score (%) on FACTS Search V2 — October 9, 2026
We mirror the published published score (%) view for FACTS Search V2. GPT-6 Astra has the highest reported value at 89.9%. These are published source configurations. Compare the task version and disclosed settings before treating a score difference as a model difference. We do not use these results to rank models overall.
GPT-6 Astra
Kaggle benchmark v2; task revision 6; Evaluated 2026-09-15
GPT-5.6 Sol
Kaggle benchmark v2; task revision 6; Evaluated 2026-07-24
GPT-6.1 Sol
Kaggle benchmark v2; task revision 6; Evaluated 2026-10-05
22 source rowsAgenticCurrentDisplay onlyUpdated October 9, 2026
Published score (%) table (22 source rows)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | GPT-6 AstraCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-09-15 | 89.9% | Not reported | Unknown |
| 2 | GPT-5.6 SolCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-24 | 88.5% | Not reported | Unknown |
| 3 | GPT-6.1 SolCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-10-05 | 88.2% | Not reported | Unknown |
| 4 | Claude Opus 5.5Creator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-09-24 | 86.5% | Not reported | Unknown |
| 5 | GPT-6 SolCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-09-29 | 86.1% | Not reported | Unknown |
| 6 | Claude Opus 5Creator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-27 | 85.3% | Not reported | Unknown |
| 7 | Gemini 4 ArgonCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-10-09 | 84.0% | Not reported | Unknown |
| 8 | Gemini 3.1 Pro PreviewCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-25 | 83.7% | Not reported | Unknown |
| 9 | GPT-5.6 TerraCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-23 | 81.7% | Not reported | Unknown |
| 10 | GPT-6 LunaCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-09-29 | 81.4% | Not reported | Unknown |
| 11 | Gemini 3 Flash PreviewCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-18 | 80.9% | Not reported | Unknown |
| 12 | Gemini 3.7 FlashCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-08-18 | 80.9% | Not reported | Unknown |
| 13 | Claude Opus 4.8Creator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-24 | 79.6% | Not reported | Unknown |
| 14 | Grok 4.5Creator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-23 | 79.4% | Not reported | Unknown |
| 15 | GPT-5.6 LunaCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-24 | 79.4% | Not reported | Unknown |
| 16 | Claude Haiku 5.5Creator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-10-09 | 77.5% | Not reported | Unknown |
| 17 | Gemini 3.8 FlashCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-09-03 | 76.3% | Not reported | Unknown |
| 18 | Gemini 3.6 FlashCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-23 | 68.2% | Not reported | Unknown |
| 19 | Gemini 3.5 FlashCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-23 | 64.3% | Not reported | Unknown |
| 20 | Gemini 3.5 Flash-LiteCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-23 | 63.4% | Not reported | Unknown |
| 21 | Gemma 4 26B A4BCreator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-28 | 60.3% | Not reported | Unknown |
| 22 | Qwen 3 235B A22B Instruct 2506Creator not reportedKaggle benchmark v2; task revision 6; Evaluated 2026-07-27 | 53.4% | Not reported | Unknown |
How to read FACTS Search V2
The current FACTS Search route serves version 2. The model can use Brave Search for up to seven agent turns, and a Gemini 3.5 Flash grader compares its answer with the question and reference answer. This board retains version 2 separately from older Search results.
The source describes an F1 factuality measure, while its download supplies separate Score and F1 fields. We display Score as published and keep every supplied metric rather than treating those fields as interchangeable. The benchmark depends on its search tool and automated grader; it does not measure every research workflow.
The complete official download includes models hidden by the source page's default filters. Source names and task versions stay separate from canonical model profiles; these results do not enter overall or category rankings.
Snapshot
These are published source configurations. Compare the task version and disclosed settings before treating a score difference as a model difference. The table retains each source configuration and its reported value.
22 source rows are shown for FACTS Search V2. The benchmark falls in the Agentic category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. FACTS Search V2 is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About FACTS Search V2
Year
2025
Tasks
Search-assisted factual questions; public and held-out sets
Format
Tool-assisted answering with an automated grader
The current FACTS Search route serves version 2. The model can use Brave Search for up to seven agent turns, and a Gemini 3.5 Flash grader compares its answer with the question and reference answer. This board retains version 2 separately from older Search results. The source describes an F1 factuality measure, while its download supplies separate Score and F1 fields. We display Score as published and keep every supplied metric rather than treating those fields as interchangeable. The benchmark depends on its search tool and automated grader; it does not measure every research workflow. The complete official download includes models hidden by the source page's default filters. Source names and task versions stay separate from canonical model profiles; these results do not enter overall or category rankings.
Freshness and provenance
Version
FACTS Search V2 2025
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does FACTS Search V2 measure?
FACTS Search V2 measures factual answering with a shared Brave Search tool. Overall, public, private, hedging, search-count, attempted-accuracy, and F1 measurements remain distinct.
Which source row has the highest reported FACTS Search V2 value?
GPT-6 Astra has the highest reported FACTS Search V2 value at 89.9%. These are published source configurations. Compare the task version and disclosed settings before treating a score difference as a model difference. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on FACTS Search V2?
The October 9, 2026 snapshot contains 22 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.