Gap narrows between open and closed AI models in cyber capabilities
Original title: The gap between open and closed AI models in cyber capabilities keeps shrinking
Open-weight models like Deepseek V4.1 Flash reach 72.8% in Vals AI cyber benchmarks at $3 per run versus $22 for closed models.
Vals AI benchmark shows frontier closed models and open models converging in cyber offense tasks. GPT-6 Sol leads Nova test at 80.8% while Deepseek scores 72.8% and completes 21 of 31 objectives in Lighthouse. Progress in open models accelerates after 100 million tokens; safety guardrails proved largely ineffective.
Unconfirmed: Long-term real-world attack prevalence remains unconfirmed
- Deepseek V4.1 Flash costs $1.19 per run versus $85 for closed models
- Gap between open and closed models in cyber capabilities now 4-7 months
- Open models can aid defenders by identifying vulnerabilities
- Organizations
- Vals AIUK AI Security Institute
- AI models
- Deepseek V4.1 FlashGPT-6 Sol
From the original article
3 min read at the-decoder.com
In the Vals AI benchmark, OpenAI's GPT-6 Sol Daybreak hits 80. 8 percent of attack objectives in simulated corporate networks. Worse, the open-weight model DeepSeek V4.
By Matthias Bastian · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.