Harmful content defect rate (Azure AI Evaluation SDK, Microsoft)
Phi-4-Reasoning-Vision-15B 1.40% leads since Mar 2026 Only in system cards GPT-4o simulates adversarial conversations with the model, and a judge flags violence, sexual, self-harm and hateful content. Scored as the share of harmful responses, lower is better.
In-house Microsoft evaluation. The tasks are not public. Variants add known jailbreaks, Tier 1 languages, or audio prompts.
- Level
- Everyday Most adults, without training
- 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.
Phi-4-Reasoning-Vision-15B leads with 1.40% among models released by Mar 2026
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 Mar 2026, and nobody has reported a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 16 more
- 1 Phi-4-Reasoning-Vision-15B 1.40%
- 2 Phi-4 Mini 3.75%
- 3 Phi-4-multimodal 4.00%
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | Phi-4-Reasoning-Vision-15B | 1.40% | Mar 2026 | Model card | |
| 2 | Phi-4 Mini | 3.75% | Mar 2025 | Model card | |
| 3 | Phi-4-multimodal | 4.00% | Mar 2025 | Model card |
2 models have held number one since its release in Mar 2025
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 |
|---|---|---|---|---|
| 4 Mar 2026 | Phi-4-Reasoning-Vision-15B technical report | 1.4 | Text to Text Safety, automated content safety defect rate, lower is better. Image to Text Safety 4.5% (Section 7 table) | Safety |
| 3 Mar 2025 | Phi-4-multimodal technical report | 4 | Defect Rate average over Violence 7%, Sexual 6%, Self-Harm 0%, Hateful 3%, lower is better (Table 10). With known jailbreaks 2.25% (Table 11). Tier 1 languages 5.06% (Table 13). Audio prompts 3.25% (Table 14) | Safety |
| 3 Mar 2025 | Phi-4 Mini technical report | 3.75 | Defect Rate average over Violence 6%, Sexual 6%, Self-Harm 0%, Hateful 3%, lower is better (Table 10). With known jailbreaks 1.25% (Table 11). Tier 1 languages average 3.91% (Table 13) | 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% - 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%