b3: Backbone Breaker Benchmark (b3 Agent Security)
We show this table for reference; we do not rank on it.
Lakera's open-source benchmark of how well the LLM inside an AI agent resists attacks such as system-prompt exfiltration, malicious code injection and tool-call manipulation. Higher means more attacks resisted.
Benchmark score on b3 Agent Security — October 29, 2026
We compile the b3 Agent Security rows from provider self-reports and secondary reports. Mistral Large 4 leads the table at 97.1%, followed by GLM-5.3 (93.0%) and Kimi K3 (88.0%). We do not use these results to rank models overall.
Mistral Large 4
Mistral
GLM-5.3
Z.AI
Kimi K3
Moonshot AI
5 modelsAgenticCurrentDisplay onlyUpdated October 29, 2026
Benchmark score table (5 models)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Mistral Large 4Mistral | 97.1% | Not reported | Open |
| 2 | GLM-5.3Z.AI | 93.0% | Not reported | Open |
| 3 | Kimi K3Moonshot AI | 88.0% | Not reported | Open |
| 4 | Gemma 4 31BGoogle | 83.0% | Not reported | Open |
| 5 | DeepSeek V4 ProDeepSeek | 59.0% | Not reported | Open |
Among the reported b3 Agent Security rows, Mistral Large 4 is first at 97.1%. The third row is 9.1 points behind. The broader top-10 range is 38.1 points, so the table still separates the published systems.
5 models have been evaluated on b3 Agent Security. The benchmark falls in the Agentic category. b3 Agent Security is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About b3 Agent Security
Year
2025
Tasks
Threat snapshots of attacks on agent backends
Format
Attack resistance rate
Difficulty
Agent security
Mistral reported these results in its October 29, 2026 Mistral Large 4 open-weight post, including the rival-model values in its chart. They are provider-run, so BenchLM shows them display-only and keeps them out of weighted scores. Rival values were published as whole percentages.
Freshness and provenance
Version
b3 Agent Security 2025
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does b3 Agent Security measure?
Lakera's open-source benchmark of how well the LLM inside an AI agent resists attacks such as system-prompt exfiltration, malicious code injection and tool-call manipulation. Higher means more attacks resisted.
Which model scores highest on b3 Agent Security?
Mistral Large 4 by Mistral currently leads with a score of 97.1% on b3 Agent Security.
How many models are evaluated on b3 Agent Security?
5 AI models have published results on b3 Agent Security in the BenchLM catalog.
Compare top models on b3 Agent Security
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 6,100+ subscribers.
One email each week. Unsubscribe anytime.