The frontier keeps moving — Sol adds an Ultra sub-agent mode and a Max reasoning tier, and ChatGPT Work is a direct shot at the agent-does-the-busywork market. Watch how the tiered pricing (Sol at $5/$30) reshapes what you run where.

OpenAI launches its new family of models with GPT-5.6

The GPT-5.6 family — Sol, Terra, and Luna — ships across ChatGPT, Codex, and the API, alongside ChatGPT Work, an agent that turns messy context into shareable artifacts.

techcrunch.com Discussion

Open weights just became a real option for running agents: GLM-5.2 is MIT-licensed, 1M-context, and within a point of Opus 4.8 on agentic coding at roughly a fifth of the cost. If you're building agents, the self-host-vs-API math just changed.

GLM-5.2 is the step change for open agents

Z.ai's MIT-licensed 753B MoE closes the gap to frontier closed models on agentic coding — and reframes what you can actually self-host.

interconnects.ai Discussion

The demand-side confirmation of the GLM story — teams are moving real workloads to open weights on cost alone. Pricing pressure on the frontier APIs is now a procurement conversation, not a benchmark footnote.

Chinese AI models are gaining ground with U.S. companies as OpenAI, Anthropic costs surge

U.S. companies are increasingly adopting Chinese open-weight models like Z.ai's GLM as inference costs from OpenAI and Anthropic surge.

cnbc.com Discussion