Standard Refusal Evaluation
o1 100% leads since December 2024
Only in system cards Models answer disallowed and benign safety-related prompts across policy categories. An autograder scores not_unsafe and not_overrefuse.
In-house OpenAI evaluation. Some cards give category scores only.
- Made by
- OpenAI
- Released
- August 2024
- Kind
- Refusing harmful requests
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- Kept private
- Strongest model tested
- o1 #117 on the capability index
- Last in a system card
- August 2025
- Last model added
- August 2025
8
Models scored in lab cards
17
System cards report it, from 1 lab
3 models share the top score of 100% among models released by February 2025
Best score per model as labs report them in their own cards. Labs pick their own settings, so these compare less cleanly than independent runs.
The newest model with a score here came out in February 2025, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- Claude Fable 5.2
-
Gemini 4 Carbon - GPT-6 Astra
- Claude Opus 5.5
- and 16 more
- 1 o1 100%
- 2 OpenAI Computer-Using Agent (CUA) 100%
- 3 o3-mini 100%
- 4 o1-preview 99.5%
- 5 o1-mini 99%
- 6 GPT-4.5 99%
- 7 OpenAI deep research 97%
- 8 GPT-4o May 2024 93%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | o1 | 100% | Dec 2024 | Lab model card | |
| 2 | OpenAI Computer-Using Agent (CUA) | 100% | Jan 2025 | Lab model card | |
| 3 | o3-mini | 100% | Jan 2025 | Lab model card | |
| 4 | o1-preview | 99.5% | Sep 2024 | Lab model card | |
| 5 | o1-mini | 99% | Sep 2024 | Lab model card | |
| 6 | GPT-4.5 | 99% | Feb 2025 | Lab model card | |
| 7 | OpenAI deep research | 97% | Feb 2025 | Lab model card | |
| 8 | GPT-4o May 2024 | 93% | Aug 2024 | Lab model card |
3 models have held number one in labs’ own numbers since its release in August 2024
Lab claims count from the day the card came out, independent runs from the model’s release.
OpenAI
Source: Labs’ model cards
OpenAI’s own number for GPT-4o passed 85% in August 2024
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Reported by the lab in its model card
Best so far
Source: Labs’ model cards
Labs report it only in system cards
We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.
17 system cards report it, from 1 lab
Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.
| Date | Document | Score | As the document writes it | Section |
|---|---|---|---|---|
| Aug 7, 2025 | GPT-5 system card | Standard Disallowed Content Evaluation, not_unsafe, 8 categories, no overall score (Table 2) | Safety | |
| Aug 7, 2025 | GPT-5 nano system card | standard disallowed content evaluation, gpt-5-thinking-nano, 8 categories, no overall score (Table 23) | Safety | |
| Aug 7, 2025 | GPT-5 mini system card | standard disallowed content evaluation, gpt-5-thinking-mini, 8 categories, no overall score (Table 23) | Safety | |
| Aug 5, 2025 | gpt-oss-20b model card | Standard Disallowed Content Evaluation, not_unsafe, 8 categories, no overall score (Table 4) | Safety | |
| Aug 5, 2025 | gpt-oss-120b model card | Standard Disallowed Content Evaluation, not_unsafe, 8 categories, no overall score (Table 4) | Safety | |
| Jul 17, 2025 | ChatGPT agent system card | Standard Disallowed Content Evaluation, not_unsafe, 8 rows from 0.988 to 1.000, no overall score (Table 1) | Safety | |
| May 16, 2025 | codex-1 system card addendum | Standard refusal evaluation, not_unsafe, 13 categories from 0.96 to 1, no overall score (Table 1) | Safety | |
| Apr 16, 2025 | o4-mini system card | Standard refusal evaluation, not_unsafe in 13 categories with no aggregate, aggregate not_overrefuse only (Table 1) | Safety | |
| Apr 16, 2025 | o3 system card | Standard refusal evaluation, not_unsafe in 13 categories with no aggregate, aggregate not_overrefuse only (Table 1) | Safety | |
| Feb 27, 2025 | GPT-4.5 system card | 99 | Standard Refusal Evaluation, not_unsafe, not_overrefuse 0.71 (Table 1) | Safety |
| Feb 25, 2025 | OpenAI deep research system card | 97 | Disallowed content evaluations, not_unsafe 0.97, not_overrefuse 0.86 (Table 3) | Safety |
| Jan 31, 2025 | o3-mini system card | 100 | Standard Refusal Evaluation, not_unsafe, not_overrefuse 0.92 (Table 1) | Safety |
| Jan 23, 2025 | OpenAI Computer-Using Agent (CUA) system card | 100 | Standard refusal evaluation, not_unsafe 100%, not_overrefuse 55% (Table 3, categories in Table 4) | Safety |
| Dec 5, 2024 | o1 system card | 100 | Standard Refusal Evaluation, not_unsafe 1 (not_overrefuse 0.93) (Table 1) | Safety |
| Sep 12, 2024 | o1-preview system card | 99.5 | Standard Refusal Evaluation, not_unsafe (not_overrefuse 0.93) (Table 1) | Safety |
| Sep 12, 2024 | o1-mini system card | 99 | Standard Refusal Evaluation, not_unsafe (not_overrefuse 0.9) (Table 1) | Safety |
| Aug 8, 2024 | GPT-4o system card | 93 | Text safety evaluations converted to audio, Audio not_unsafe (Table 5, not_overrefuse 0.82); Appendix A Table 11 gives New GPT-4o Audio not_unsafe 1.0, not_overrefuse 0.91 | Safety |
More in Safety and alignment
- FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
- StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is.
Inkling 98.6% - Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. Gemini 4 Carbon 0%
- DistressBench In use Models reply as a chatbot to 718 simulated users in psychological crisis, in single-turn and multi-turn chats. Three judge models score replies against clinician-written rubrics.
Muse Spark 1.2 88.4% - Lakera B3 (Backbone Breaker Benchmark) In use Crowdsourced prompt-injection attacks hit backbone LLMs inside agent threat snapshots at three defense levels. The score is attack resistance, 100 minus the board's vulnerability percentage. Claude Opus 4.5 89%
- IHEval In use Models face conflicts between system, user and tool instructions. Reports a composite score for following the higher-priority instruction. GPT-4o 70%