Z.ai, Google push open LLMs and stacks for production agents

Z.ai released GLM‑4.7 on December 22, 2025. The model is publicly available and tuned for extended, multi-step coding and tool-use workflows. It supports a ~202.8k-token context window (Z.ai blog/docs).

Google rolled out Gemini 3 Flash on December 17, 2025. Gemini 3 Flash is a lower-cost, higher-throughput variant that Google is making the default in the Gemini app and exposing via the Gemini CLI and Vertex for developer and enterprise workflows (Google blog; Google Developers).

Concurrent vendor moves also surfaced this week; for example, OpenAI Skills in Codex aims to boost agent efficiency.

Why It Matters

• Engineers can run longer, stateful agent sessions without constant retrieval by using models with very large context windows (GLM‑4.7: ~202.8k tokens).
• Gemini 3 Flash cuts per-task latency and cost, making high-frequency agent orchestration economically practical (Google: <¼ the cost of Gemini 3 Pro in some previews).
• Open, deployable weights and CLI/tooling let teams pick self-hosted or cloud paths for compliance, control, and lower latency (GLM‑4.7 weights on Hugging Face; Gemini available via CLI and Vertex).
• Improved tool-calling stability and agentic modes should shorten integration cycles, reduce prompt engineering, and increase reproducibility in production pipelines.