Focused ASHFOG writing connected by this subject.
Cloudflare's security-audit-skill turns coding agents into a structured audit pipeline with coverage tracking, adversarial validation, and machine-readable findings.
Z.ai's GLM-5.3 rollout targets stronger coding, long-horizon agent work, and vulnerability discovery, while its public-weight release follows a more cautious security path.
Moonshot AI's 2.8-trillion-parameter Kimi K3 delivers serious coding, agentic and multimodal capability, while benchmarks, latency and deployment costs complicate the hype.
DeepSeek's official V4 Flash release sharply improves coding and tool-use scores, adds DSpark decoding, and reveals the trade-offs behind its agent-first design.
