Meta Muse Glimmer 30B Card - Siren AgentDojo - Attack Success Rate: leaderboard
AgentDojo tool-use tasks (949 injection scenarios) attacked by an adaptive LLM attacker (Claude Opus 4.6, up to 6 iterations); success rate of the injected malicious task. Meta Muse Glimmer 30B model card (August 2026); the methodology report gives internal runs for all three models, Glimmer at high reasoning and Gemma 4 31B and Qwen3.6-27B in thinking mode; percent, lower is better.
Metric: Attack Success Rate (%, lower is better). Source: huggingface.co. Saturation forecast: Around February 2027. 3 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemma 4 31B (Reasoning) | 25.6 | #212 (Gemma 4 31B) |
| 2 | Muse Glimmer 30B (High) | 28.4 | #191 (Muse Glimmer 30B) |
| 3 | Qwen 3.6 27B (Reasoning) | 40.3 | #182 (Qwen 3.6 27B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=meta-muse-glimmer-30b-card-siren-agentdojo-attack-success-rate · How It Works · Data refreshed daily, snapshot 2026-11-05.