Mistral Large 4 Release - B3 Agent Security - Attack Resistance: leaderboard
Mistral Large 4 open-weight release post (2026-10-29): B3 Agent Security Benchmark attack resistance, the share of attacks such as system-prompt exfiltration or malicious code injection that the model resisted, as reported by Mistral for six models. Reasoning effort not stated. Attack resistance (%).
Metric: Attack Resistance (%). Source: mistral.ai. Saturation forecast: Estimated already saturated. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GLM-5.3 | 93.3 | #44 |
| 2 | Mistral Large 4 | 93.3 | #106 |
| 3 | Kimi K3 | 88.1 | #35 |
| 4 | Gemma 4 31B | 82.9 | #212 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=mistral-large-4-release-b3-agent-security-attack-resistance · How It Works · Data refreshed daily, snapshot 2026-11-05.