FACTS Benchmark Suite

We show this table for reference; we do not rank on it.

Data verified October 13, 2026

The official FACTS suite combines document grounding, visual factuality, parametric recall, and search-assisted answering. The published average stays separate from its component scores.

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.

Should you trust a FACTS Benchmark Suite score?

No caution flags. Scores are separated; the table is mirrored and not verified row by row; the benchmark is current.

Saturation Different source configurations: These are published source configurations. Compare the task version and disclosed settings before treating a score difference as a model difference.
Who produced the rows Mirrored table: Every row is mirrored from Google DeepMind, Google Research, and Kaggle as it publishes it; BenchLM did not re-run these results. Source.
Freshness and weight Display only: Current, refreshed quarterly. Display only; it does not move rankings. Public benchmark set.

Row statuses come from the verification registry; saturation and freshness from the benchmark lifecycle. A provider self-report is displayable evidence, not an independent run.

Published average (%) on FACTS Benchmark Suite — October 13, 2026

We mirror the published published average (%) view for FACTS Benchmark Suite. GPT-6 Astra has the highest reported value at 75.7%. 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 v1; task revision 1; 4/4 component scores; Evaluated 2026-09-15

75.7%

GPT-6.1 Sol

Kaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-10-13

75.3%

GPT-6 Sol

Kaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-09-29

72.5%

40 source rowsKnowledgeCurrentDisplay onlyUpdated October 13, 2026

Published average (%) table (40 source rows)

Published average (%) results for FACTS Benchmark Suite
Rank	Model / configuration	Score	Parameters (B)	Open / closed
1	GPT-6 AstraCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-09-15	75.7%	Not reported	Unknown
2	GPT-6.1 SolCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-10-13	75.3%	Not reported	Unknown
3	GPT-6 SolCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-09-29	72.5%	Not reported	Unknown
4	Gemini 3.7 FlashCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-08-27	71.2%	Not reported	Unknown
5	Gemini 3.8 FlashCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-09-04	69.4%	Not reported	Unknown
6	Gemini 3.1 Pro PreviewCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-07-25	67.4%	Not reported	Unknown
7	GPT-5.5Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluation date unreported	66.7%	Not reported	Unknown
8	Gemini 3.6 FlashCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-07-23	65.3%	Not reported	Unknown
9	Claude Opus 5.5Creator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-09-25	65.3%	Not reported	Unknown
10	Gemini 3 Flash PreviewCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-07-18	63.4%	Not reported	Unknown
11	GPT-6 LunaCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-09-29	63.3%	Not reported	Unknown
12	Gemini 3.5 FlashCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-07-23	63.1%	Not reported	Unknown
13	Gemini 2.5 ProCreator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	61.5%	Not reported	Unknown
14	GPT-5Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	56.5%	Not reported	Unknown
15	Claude Opus 4.8Creator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-07-24	53.5%	Not reported	Unknown
16	Gemini 3.5 Flash-LiteCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-07-23	53.5%	Not reported	Unknown
17	GPT-5.6 SolCreator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2026-07-24	52.4%	Not reported	Unknown
18	GPT-5.6 TerraCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-07-24	52.0%	Not reported	Unknown
19	Gemini 3.1 Flash-Lite PreviewCreator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluation date unreported	51.4%	Not reported	Unknown
20	Grok 4.5Creator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-07-28	50.2%	Not reported	Unknown
21	GPT-5.2Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluation date unreported	48.5%	Not reported	Unknown
22	Gemini 2.5 FlashCreator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	47.2%	Not reported	Unknown
23	Grok 4Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	46.3%	Not reported	Unknown
24	GPT-4.1Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	45.8%	Not reported	Unknown
25	Claude Opus 4.7Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluation date unreported	45.5%	Not reported	Unknown
26	Gemma 4 31BCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-07-27	45.3%	Not reported	Unknown
27	GPT-5.1Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	45.0%	Not reported	Unknown
28	o3Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	44.4%	Not reported	Unknown
29	Claude Opus 4.5Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	43.9%	Not reported	Unknown
30	Gemma 4 26B A4BCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-07-28	42.7%	Not reported	Unknown
31	Claude Opus 4.6Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluation date unreported	41.5%	Not reported	Unknown
32	Claude Sonnet 4.5 (thinking)Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	40.6%	Not reported	Unknown
33	Claude Opus 4.1Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	40.4%	Not reported	Unknown
34	GPT-5 miniCreator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	38.6%	Not reported	Unknown
35	Claude Sonnet 4.6Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluation date unreported	37.0%	Not reported	Unknown
36	GPT-5.6 LunaCreator not reportedKaggle benchmark v1; task revision 1; 4/4 component scores; Evaluated 2026-08-12	35.0%	Not reported	Unknown
37	Claude Sonnet 4Creator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	35.0%	Not reported	Unknown
38	Gemma 3 27BCreator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	29.1%	Not reported	Unknown
39	o4 miniCreator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	28.1%	Not reported	Unknown
40	grok-4-fast-reasoningCreator not reportedKaggle benchmark v1; task revision 1; 3/4 component scores; Evaluated 2025-12-07	25.5%	Not reported	Unknown

How to read FACTS Benchmark Suite

Some rows lack a component result even though the table labels the average as four dimensions. The retained measurements preserve those missing cells. A published average does not establish that a model completed all four tests, and it should be read alongside its coverage.

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

40 source rowsBenchmark v1Complete downloadDisplay only

Official Kaggle leaderboard Official leaderboard JSON Evaluation method Benchmark-page Apache 2.0 declaration FACTS suite introduction

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.

40 source rows are shown for FACTS Benchmark Suite. The benchmark falls in the Knowledge category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. FACTS Benchmark Suite is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.

About FACTS Benchmark Suite

Year

2025

Tasks

Four benchmark dimensions; public and held-out examples

Format

Published percentage average

The FACTS suite brings four factuality tests together: Grounding v2, Multimodal v1, Parametric v1, and Search v2 in this source snapshot. We show Kaggle's published Average without recalculating it or borrowing scores from provider reports. Some rows lack a component result even though the table labels the average as four dimensions. The retained measurements preserve those missing cells. A published average does not establish that a model completed all four tests, and it should be read alongside its coverage. 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.

FACTS Benchmark Suite methodology Public benchmark source

Freshness and provenance

Version

FACTS Benchmark Suite 2025

Refresh cadence

Quarterly

Staleness state

Current

Question availability

Public benchmark set

CurrentDisplay only

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 Benchmark Suite measure?

The official FACTS suite combines document grounding, visual factuality, parametric recall, and search-assisted answering. The published average stays separate from its component scores.

Which source row has the highest reported FACTS Benchmark Suite value?

GPT-6 Astra has the highest reported FACTS Benchmark Suite value at 75.7%. 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 Benchmark Suite?

The October 13, 2026 snapshot contains 40 AI models.

Last updated: October 13, 2026 · mirrored from the public benchmark leaderboard

Know when it’s worth switching models

The model to choose, the cheaper alternative, and the release we would wait on.

Read a sample issue

Join 6,100+ subscribers.

One email each week. Unsubscribe anytime.