All Benchmarks/Factuality

FACTS Multimodal

Higher is better · Image-input models only

Answer questions about images factually, combining what is in the image with general knowledge. An answer counts only if it includes the essential facts and contradicts nothing. Only models that accept images can take part. Scores run from 0 to 100%. Higher is better.

Other versions:FACTS Benchmark SuiteFACTS GroundingFACTS ParametricFACTS Search
Top score50.7%Gemini 3.8 Flash
Models tested31
Last updated2026.09.25
Model release date (newest on the right)
Best score so farHigher on the chart is better
  1. 1
    Google
    Gemini 3.8 Flash
    2026.09.03
    50.7%
  2. 2
    Google
    Gemini 3.7 Flash
    2026.08.13
    49.8%
  3. 3
    Anthropic
    Claude Opus 5.5
    2026.09.22
    48.4%
  4. 4
    OpenAI
    GPT-6 Astra
    2026.09.04
    47.7%
  5. 5
    Google
    Gemini 3.6 Flash
    2026.07.21
    47.2%
  6. 6
    Google
    Gemini 2.5 Pro
    2025.06.17
    46.9%
  7. 7
    OpenAI
    GPT-5.5
    2026.04.23
    45.8%
  8. 8
    OpenAI
    GPT-5
    2025.08.07
    44.1%
  9. 9
    OpenAI
    GPT-6 Sol
    2026.09.22
    43.1%
  10. 10
    OpenAI
    GPT-5 Mini
    2025.08.07
    41.5%

The "harness" is the agent program the AI used to carry out the task. The same model can score very differently depending on its harness and reasoning effort.