We Benchmarked Frontier LLMs on Defensive Security. The Results Surprised Us

cotool.ai

6 pointslogancarmody2 comments

Interesting. I wonder if Gemini 3 reverses that performance trend or if the agent harness lended itself to OpenAI / Anthropic more than Google.

Would like to see this on more open-source agent harnesses and tools.

agreed - would be interesting