Gpt-5.4-mini

Microsoft’s MDASH adds a cybersecurity-specific model (MAI-Cyber-1-Flash with GPT-5.4) that scores 95.95% on CyberGym and claims a 50% lower configuration cost than the existing MDASH stack.

Five frontier LLMs were tested on 20 real CVEs across three prompt types; no model reliably fixes vulnerabilities, with a best 50% solve rate and significant cross-family differences; token cost varies up to ~4x by model, and locate prompts are the hardest test of genuine security reasoning.

Interfaze markets a CNN/DNN+transformer hybrid architecture claiming high deterministic-task accuracy at scale, backed by benchmark leadership, large context, and competitive pricing.