BlogLLM Benchmark Scores as Published, Side by Side
OpenAI deception evals: Results as Published
Every result the catalog holds for OpenAI deception evals: 3 models from 1 developer, read from 1 page. Each number is as printed by the company named beside it; nothing is ranked or rescored. Models are listed by developer, newest first.
Area Safety and AlignmentMaintainer OpenAIIntroduced 2025Data version 2026103001, as of 2026-10-30
OpenAI internal evaluations of deception: settings with impossible coding tasks, missing multimodal assets and a sample of production traffic, in which a response counts as deceptive when the model claims to have done what it could not; the chart prints the deception rate across these settings, so lower is better.
Not checked against a maintainer's page.
OpenAI deception evals (deception rate)
lower is better
| Model | Value | Printed by | Setup as printed |
|---|---|---|---|
| GPT-5OpenAI, released 2025-08-07 | 4.8% | OpenAI, Introducing GPT-5checked 2026-10-15 | Effort: high |
| 11.6% | OpenAI, Introducing GPT-5checked 2026-10-15 | Effort: none | |
| o3OpenAI, released 2025-04-16 | 22% | OpenAI, Introducing GPT-5checked 2026-10-15 | Effort: high |
| GPT-4oOpenAI, released 2024-05-13 | 20.6% | OpenAI, Introducing GPT-5checked 2026-10-15 | - |
Every number here is read from the same catalog file as the benchmark table, data version 2026103001, and is the number the table shows for that model and test. Where one page printed a model under several setups, each value is listed. The catalog is public in armanobosyan/sugra-llm-benchmarks; a wrong number can be reported from the table, with Report an error under the value, or as an issue in that repository.