FACTS Benchmark Suite leaderboard
- GPT-6 Astra leads
- 75.7%
- models measured
- 37
- lead over #2
- +0.4 pts
- median
- 51.4%
Leader on FACTS Benchmark Suite · read October 21
On October 21, 2026, 37 of our catalog's models had a FACTS Benchmark Suite result, and OpenAI GPT-6 Astra led with 75.7%, ahead of OpenAI GPT-6.1 Sol at 75.3%.
A benchmark suite from Google DeepMind and Google Research that evaluates the factual accuracy and grounding of models. It has four parts: answers grounded in the context of a prompt, questions about input images, factoid questions answered from the model's own knowledge, and questions answered by using search as a tool. The number is the average accuracy across the four parts, as Kaggle publishes it.
Role in the index
FACTS Benchmark Suite is shown as published and does not feed any of our indexes. Read the methodology
| # | Model | Result | Evidence | Coding index | Output $/Mtok |
|---|---|---|---|---|---|
| 1 | GPT-6 AstraOpenAI | 75.7% | 85.1 | $50.00 | |
| 2 | GPT-6.1 SolOpenAI | 75.3% | 82.4 | $10.00 | |
| 3 | GPT-6 SolOpenAI | 72.5% | 76.0 | $10.00 | |
| 4 | Gemini 3.7 FlashGoogle | 71.2% | 58.0 | $3.75 | |
| 5 | Gemini 3.8 FlashGoogle | 69.4% | 61.6 | $3.75 | |
| 6 | Gemini 3.1 Pro PreviewGoogle | 67.4% | 32.3 | $12.00 | |
| 7 | GPT-5.5OpenAI | 66.7% | 71.3 | $30.00 | |
| 8 | Gemini 3.6 FlashGoogle | 65.3% | 62.5 | $3.75 | |
| 9 | Claude Opus 5.5Anthropic | 65.3% | 88.8 | $20.00 | |
| 10 | Gemini 3 Flash PreviewGoogle | 63.4% | 43.7 | $3.00 | |
| 11 | GPT-6 LunaOpenAI | 63.3% | 57.3 | $0.50 | |
| 12 | Gemini 3.5 FlashGoogle | 63.1% | 39.8 | $9.00 | |
| 13 | Gemini 2.5 ProGoogle | 61.5% | 19.1 | $10.00 | |
| 14 | GPT-5OpenAI | 56.5% | 38.8 | $10.00 | |
| 15 | Claude Opus 4.8Anthropic | 53.5% | 65.0 | $25.00 | |
| 16 | Gemini 3.5 Flash LiteGoogle | 53.5% | 49.0 | $2.50 | |
| 17 | GPT-5.6 SolOpenAI | 52.4% | 73.5 | $20.00 | |
| 18 | GPT-5.6 TerraOpenAI | 52.0% | 63.9 | $12.00 | |
| 19 | Gemini 3.1 Flash Lite PreviewGoogle | 51.4% | 22.7 | $1.50 | |
| 20 | Grok 4.5xAI | 50.2% | 58.6 | $6.00 | |
| 21 | GPT-5.2OpenAI | 48.5% | 55.9 | $14.00 | |
| 22 | Gemini 2.5 FlashGoogle | 47.2% | Not benchmarked | $2.50 | |
| 23 | GPT-4.1OpenAI | 45.8% | Not benchmarked | $8.00 | |
| 24 | Claude Opus 4.7Anthropic | 45.5% | 67.5 | $25.00 | |
| 25 | Gemma 4 31BGoogle | 45.3% | 39.0 | $0.40 | |
| 26 | GPT-5.1OpenAI | 45.0% | 40.4 | $10.00 | |
| 27 | o3OpenAI | 44.4% | Not benchmarked | $8.00 | |
| 28 | Claude Opus 4.5Anthropic | 43.9% | 44.0 | $25.00 | |
| 29 | Gemma 4 26B A4BGoogle | 42.7% | Not benchmarked | $0.60 | |
| 30 | Claude Opus 4.6Anthropic | 41.5% | 57.2 | $25.00 | |
| 31 | Claude Sonnet 4.5Anthropic | 40.6% | 39.5 | $15.00 | |
| 32 | Claude Opus 4.1Anthropic | 40.4% | Not benchmarked | $75.00 | |
| 33 | GPT-5 MiniOpenAI | 38.6% | 34.2 | $2.00 | |
| 34 | Claude Sonnet 4.6Anthropic | 37.0% | 57.3 | $15.00 | |
| 35 | GPT-5.6 LunaOpenAI | 35.0% | 58.0 | $1.20 | |
| 36 | Gemma 3 27BGoogle | 29.1% | Not benchmarked | $0.30 | |
| 37 | o4 MiniOpenAI | 28.1% | Not benchmarked | $4.40 |
Only models in our catalog are listed, and each appears once with its best published run. 59 rows were read from the source on October 21, 2026, and rows that match no catalog model are left out.
| Model | Result | Grounding | Multimodal | Parametric | Search |
|---|---|---|---|---|---|
| OpenAI GPT-6 Astra | 75.7% | 82.0% | 47.7% | 83.1% | 89.9% |
| OpenAI GPT-6.1 Sol | 75.3% | 82.4% | 48.5% | 82.1% | 88.2% |
| OpenAI GPT-6 Sol | 72.5% | 85.2% | 43.1% | 75.6% | 86.1% |
| Google Gemini 3.7 Flash | 71.2% | 75.5% | 49.8% | 78.7% | 80.9% |
| Google Gemini 3.8 Flash | 69.4% | 73.1% | 50.7% | 77.3% | 76.3% |
| Google Gemini 3.1 Pro Preview | 67.4% | 65.7% | 41.3% | 79.0% | 83.7% |
| OpenAI GPT-5.5 | 66.7% | 76.2% | 45.8% | 78.0% | Not published |
| Google Gemini 3.6 Flash | 65.3% | 72.5% | 47.2% | 73.4% | 68.2% |
| Anthropic Claude Opus 5.5 | 65.3% | 68.5% | 48.4% | 57.9% | 86.5% |
| Google Gemini 3 Flash Preview | 63.4% | 59.0% | 41.3% | 72.3% | 80.9% |
| OpenAI GPT-6 Luna | 63.3% | 81.4% | 34.2% | 56.1% | 81.4% |
| Google Gemini 3.5 Flash | 63.1% | 73.2% | 41.4% | 73.5% | 64.3% |
| Google Gemini 2.5 Pro | 61.5% | 74.3% | 46.9% | 63.2% | Not published |
| OpenAI GPT-5 | 56.5% | 69.6% | 44.1% | 55.8% | Not published |
| Anthropic Claude Opus 4.8 | 53.5% | 69.7% | 39.8% | 25.1% | 79.6% |
| Google Gemini 3.5 Flash Lite | 53.5% | 71.2% | 35.8% | 43.4% | 63.4% |
| OpenAI GPT-5.6 Sol | 52.4% | 68.7% | 0.0% | Not published | 88.5% |
| OpenAI GPT-5.6 Terra | 52.0% | 71.2% | 0.0% | 55.0% | 81.7% |
| Google Gemini 3.1 Flash Lite Preview | 51.4% | 66.5% | 39.4% | 48.4% | Not published |
| SpaceXAI Grok 4.5 | 50.2% | 74.5% | 0.0% | 46.8% | 79.4% |
| OpenAI GPT-5.2 | 48.5% | 76.2% | 39.7% | 29.7% | Not published |
| Google Gemini 2.5 Flash | 47.2% | 70.0% | 41.0% | 30.7% | Not published |
| OpenAI GPT-4.1 | 45.8% | 45.6% | 40.1% | 51.5% | Not published |
| Anthropic Claude Opus 4.7 | 45.5% | 61.7% | 35.8% | 38.8% | Not published |
| Google Gemma 4 31B | 45.3% | 80.7% | 24.7% | 7.8% | 68.1% |
| OpenAI GPT-5.1 | 45.0% | 50.0% | 41.8% | 43.2% | Not published |
| OpenAI o3 | 44.4% | 36.2% | 39.9% | 57.1% | Not published |
| Anthropic Claude Opus 4.5 | 43.9% | 62.1% | 39.2% | 30.6% | Not published |
| Google Gemma 4 26B A4B | 42.7% | 80.9% | 23.9% | 5.7% | 60.3% |
| Anthropic Claude Opus 4.6 | 41.5% | 54.9% | 36.8% | 33.0% | Not published |
| Anthropic Claude Sonnet 4.5 | 40.6% | 61.8% | 31.1% | 29.0% | Not published |
| Anthropic Claude Opus 4.1 | 40.4% | 54.8% | 33.1% | 33.2% | Not published |
| OpenAI GPT-5 Mini | 38.6% | 58.3% | 41.5% | 16.0% | Not published |
| Anthropic Claude Sonnet 4.6 | 37.0% | 56.2% | 32.6% | 22.1% | Not published |
| OpenAI GPT-5.6 Luna | 35.0% | 0.0% | 35.8% | 25.0% | 79.4% |
| Google Gemma 3 27B | 29.1% | 56.4% | 24.6% | 6.3% | Not published |
| OpenAI o4 Mini | 28.1% | 29.3% | 34.5% | 20.5% | Not published |
The parts as Kaggle Benchmarks publishes them, for the same run as the result.