Single-turn benign request evaluations (Anthropic)
Claude Fable 5.1 0% leads since September 2026
Only in system cards Models answer benign requests on sensitive topics. Score is the refusal rate, where lower is better.
In-house Anthropic evaluation. The tasks are not public.
- Made by
- Anthropic
- Released
- May 2025
- Kind
- Refusing harmful requests
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- Kept private
- Strongest model tested
- Claude Fable 5.2 #1 on the capability index
- Last in a system card
- November 2026
- Last model added
- November 2026
19
Models scored in lab cards
19
System cards report it, from 1 lab
2 models share the top score of 0%
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.
- 1 Claude Fable 5.1 0%
- 2 Claude Fable 5.2 0%
- 3 Claude Fable 5 0.01%
- 4 Claude Sonnet 4.5 0.02%
- 5 Claude Haiku 4.5 0.02%
- 6 Claude Sonnet 5.5 0.02%
- 7 Claude Opus 5.5 0.03%
- 8 Claude Mythos Preview 0.06%
- 9 Claude Opus 4 0.07%
- 10 Claude Opus 4.1 0.08%
- 11 Claude Opus 5 0.09%
- 12 Claude Haiku 5.5 0.17%
- 13 Claude Sonnet 4 0.23%
- 14 Claude Opus 4.5 0.23%
- 15 Claude Opus 4.7 0.28%
- 16 Claude Opus 4.8 0.36%
- 17 Claude Sonnet 4.6 0.41%
- 18 Claude Sonnet 5 0.59%
- 19 Claude Opus 4.6 0.68%
Source: Labs’ model cards
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Claude Fable 5.1 | 0% | Sep 2026 | Lab model card | |
| 2 | Claude Fable 5.2 | 0% | Nov 2026 | Lab model card | |
| 3 | Claude Fable 5 | 0.01% | Jun 2026 | Lab model card | |
| 4 | Claude Sonnet 4.5 | 0.02% | Sep 2025 | Lab model card | |
| 5 | Claude Haiku 4.5 | 0.02% | Oct 2025 | Lab model card | |
| 6 | Claude Sonnet 5.5 | 0.02% | Sep 2026 | Lab model card | |
| 7 | Claude Opus 5.5 | 0.03% | Sep 2026 | Lab model card | |
| 8 | Claude Mythos Preview | 0.06% | Apr 2026 | Lab model card | |
| 9 | Claude Opus 4 | 0.07% | May 2025 | Lab model card | |
| 10 | Claude Opus 4.1 | 0.08% | Aug 2025 | Lab model card | |
| 11 | Claude Opus 5 | 0.09% | Jul 2026 | Lab model card | |
| 12 | Claude Haiku 5.5 | 0.17% | Oct 2026 | Lab model card | |
| 13 | Claude Sonnet 4 | 0.23% | May 2025 | Lab model card | |
| 14 | Claude Opus 4.5 | 0.23% | Nov 2025 | Lab model card | |
| 15 | Claude Opus 4.7 | 0.28% | Apr 2026 | Lab model card | |
| 16 | Claude Opus 4.8 | 0.36% | May 2026 | Lab model card | |
| 17 | Claude Sonnet 4.6 | 0.41% | Feb 2026 | Lab model card | |
| 18 | Claude Sonnet 5 | 0.59% | Jun 2026 | Lab model card | |
| 19 | Claude Opus 4.6 | 0.68% | Feb 2026 | Lab model card |
4 models have held number one in labs’ own numbers since its release in May 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
Anthropic
Source: Labs’ model cards
Scores over time
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.
19 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 |
|---|---|---|---|---|
| Nov 9, 2026 | Claude Fable 5.2 system card | 0 | Single-turn benign request evaluations, overall refusal rate, API without a system prompt. Claude.ai 0.19%, lower is better (Table 4.1.2.A) | Safety |
| Oct 7, 2026 | Claude Haiku 5.5 system card | 0.17 | Single-turn benign request evaluations, overall refusal rate, API without system prompt (claude.ai 0.82%) (Table 4.1.2.A) | Safety |
| Sep 28, 2026 | Claude Sonnet 5.5 system card | 0.02 | Single-turn benign requests, overall refusal rate, API without system prompt (claude.ai 0.20%), lower is better (Table 4.1.2.A) | Safety |
| Sep 22, 2026 | Claude Opus 5.5 system card | 0.03 | Single-turn benign requests, overall over-refusal rate, API without a system prompt, lower is better (claude.ai 0.38%) (Table 4.1.2.A) | Safety |
| Sep 1, 2026 | Claude Fable 5.1 system card | 0 | Single-turn benign request evaluations, overall refusal rate, API without a system prompt; claude.ai 0.34%, lower is better (Table 4.1.2.A) | Safety |
| Jul 24, 2026 | Claude Opus 5 system card | 0.09 | Single-turn benign requests, overall over-refusal rate, API without a system prompt, lower is better (claude.ai 0.47%) (Table 4.1.2.A) | Safety |
| Jun 30, 2026 | Claude Sonnet 5 system card | 0.59 | Single-turn benign requests, overall refusal rate, API without a system prompt, lower is better (claude.ai 1.54%) (Table 4.1.2.A) | Safety |
| Jun 9, 2026 | Claude Fable 5 system card | 0.01 | Single-turn benign request evaluations, overall refusal rate, API without a system prompt; claude.ai 0.49%, lower is better (Table 4.1.2.A) | Safety |
| May 28, 2026 | Claude Opus 4.8 system card | 0.36 | Single-turn benign requests, overall over-refusal rate, API without a system prompt, lower is better (claude.ai 0.49%) (Table 4.1.2.A) | Safety |
| Apr 16, 2026 | Claude Opus 4.7 system card | 0.28 | Single-turn benign request evaluations, overall refusal rate, lower is better. Without thinking 0.50, with adaptive thinking 0.06 (Table 4.1.2.A) | Safety |
| Apr 7, 2026 | Claude Mythos Preview system card | 0.06 | Single-turn benign request evaluations, overall refusal rate, lower is better (Table 8.1.1.2.A); 0.09 without thinking, 0.02 with thinking | Safety |
| Feb 17, 2026 | Claude Sonnet 4.6 system card | 0.41 | Single-turn benign request evaluations, overall refusal rate, lower is better (Table 3.1.2.A) | Safety |
| Feb 5, 2026 | Claude Opus 4.6 system card | 0.68 | Single-turn benign request evaluations, overall refusal rate, lower is better, 7 languages (Table 3.1.2.A); default 0.80, extended thinking 0.56 | Safety |
| Nov 24, 2025 | Claude Opus 4.5 system card | 0.23 | Single-turn benign request evaluations, overall refusal rate, lower is better, 6 languages (Table 3.1.2.A; default 0.18%, extended thinking 0.27%) | Safety |
| Oct 15, 2025 | Claude Haiku 4.5 system card | 0.02 | Single-turn benign request evaluations, overall refusal rate, lower is better (Table 2.1.2.A) | Safety |
| Sep 29, 2025 | Claude Sonnet 4.5 system card | 0.02 | Single-turn benign request evaluations, overall refusal rate 0.02% (± 0.04%), lower is better, run without safety classifiers. Standard thinking 0.05%, extended thinking 0.00% (Table 2.1.2.A) | Safety |
| Aug 5, 2025 | Claude Opus 4.1 system card addendum | 0.08 | Single-turn benign request evaluations, overall refusal rate 0.08% (± 0.09%), lower is better. Standard thinking 0.13%, extended thinking 0.04% (Table 2.1.B) | Safety |
| May 22, 2025 | Claude Sonnet 4 system card | 0.23 | Single-turn benign request evaluations, overall refusal rate 0.23% (± 0.11%), lower is better. Standard thinking 0.43%, extended thinking 0.03% (Table 2.2.A) | Safety |
| May 22, 2025 | Claude Opus 4 system card | 0.07 | Single-turn benign request evaluations, overall refusal rate 0.07% (± 0.07%), lower is better. Standard thinking 0.12%, extended thinking 0.02% (Table 2.2.A) | 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%