Skip to content
TechIndexSubmit
Back to directory

open-llm-safety-gauntlet

Six open LLMs, three attack axes, fully local — measuring attack-resistance vs over-refusal, and how far to trust a local judge.

Similar tools

  • 945-generation adversarial evaluation of 3 open LLMs across 3 personas and 20 attack types, measuring semantic drift, override rates, and distributional instability.

    Models
  • Bilingual (Polish + English) quality evals for local open LLMs as agents and assistants: tool calling, multi-step tasks, frozen-web shopping, instruction following, writing. Every result paired with a replicable run spec.

    Models
  • First-of-its-kind AI benchmark for evaluating the protection capabilities of large language model (LLM) guard systems (guardrails and safeguards)

    Models
  • Empirical LLM evaluation on CPU-only testlab hardware — 5 models benchmarked for agent-capability: speed, tool-call validity, and reasoning quality

    Models
  • Assessing the reliability, accuracy, and safety of large language models, RAG pipelines, and AI agents.

    Models
  • Open-source LLM inference benchmarks — TTFT, TPOT, Throughput, Latency & Cost-per-token for models like Llama, Qwen, Gemma, DeepSeek, Gpt-Oss etc. deployed on different dedicated GPUs.

    Models