CyberBench (Liu et al.)
Muse Wonder Full 0.80% leads since Oct 2026 Only in system cards Models face detail-seeking high-severity cyber requests. Meta uses it as a seed set for adaptive red-teaming and reports attack success rate.
AAAI-24 AICS workshop benchmark. Not the same as Cybench. The original is a 10-dataset cyber NLP suite (NER, summarization, multiple choice, classification). Meta describes and uses it as a source of harmful cyber requests.
- Level
- Expert Specialists with years of training, often over hours of work
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Open-ended Free-form text, documents, designs or media judged as a whole.
Muse Wonder Full leads with 0.80%
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 Muse Wonder Full 0.80%
- 2 Muse Spark 1.3 6.30%
- 3 Muse Spark 1.1 9.00%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Muse Wonder Full | 0.80% | Oct 2026 | Model card | |
| 2 | Muse Spark 1.3 | 6.30% | Oct 2026 | Model card | |
| 3 | Muse Spark 1.1 | 9.00% | Jul 2026 | Model card |
2 models have held number one since its release in Feb 2024
Lab claims count from the day the card came out, independent runs from the model’s release.
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Labs report it only in system cards
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark. Some system cards do, below.
3 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 |
|---|---|---|---|---|
| 23 Oct 2026 | Muse Spark 1.3 system card | 6.3 | CyberBench multi-turn ASR, Muse Spark 1.3 (API) (Figure 14). Single-turn 2.8 | Cyber risk |
| 12 Oct 2026 | Muse Wonder Full model card | 0.8 | CyberBench multi-turn ASR, lower is better (Cyber Robustness Evaluations table, section 4). Single turn 0.8 | Cyber risk |
| 9 Jul 2026 | Muse Spark 1.1 system card | 9 | CyberBench multi-turn ASR, Muse Spark 1.1 model-only (Figure 19 text). Single-turn 3.0. System 0.8 and 4.6 | Cyber risk |
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% - AgentDojo In use Agents do tool-use tasks while tools and data carry prompt injections. Attack success rate counts hijacked runs. Grok 4 Fast 0%
- MASK In use Measures whether a model contradicts its own elicited beliefs under pressure to lie. Honesty separated from accuracy. Muse Spark Open 98.1%
- 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. GPT-6 Astra 8.50%
- Agentic misalignment suite (Anthropic) In use Agents in simulated company settings face threats or goal conflicts that invite blackmail, leaking or sabotage. Score is the misaligned action rate. Claude Sonnet 4.5 0%