GPT-5.1
OpenAI · Released 2025-11-13 · updated 11/03 20:05
Each has its strengths; see them more clearly.
Each capability is calculated separately. Without enough real tests, it stays blank.
Category scores reflect ranking support within each reference group. They cannot be added together or used to compare absolute levels across capabilities.
How stable is the overall rank?
Remove one benchmark or organization at a time and adjust individual weights and error handling to see how ranks shift.
4 scenarios lack enough evidence to rank.Rank 60–79 with the original candidates unchanged. This range is not a confidence interval and does not include results never published.
Every result has a source.
These are the model's public aggregate results. Expand to view run configuration and how each result is used.
综合评测与体验4
推理2
知识2
视觉理解1
Scored benchmarks not tested
These benchmarks haven't published this model's results. Missing benchmarks score no zero.
Some unknowns remain
Missing benchmarks are never scored zero. Ranks change with new evidence, and small gaps between close scores shouldn't be over-read.
How it's calculated →