Z-AI's latest flagship for long-horizon tasks. A substantial leap over GLM-5.1, delivering a solid 1M-token context for the first time.

Efficiency-focused MoE — 284B total / 13B active with a 1M-token window. Tuned for fast, high-throughput inference while holding up on reasoning and coding.

MoE with 1.6T total parameters (49B active) and a 1M-token context. Built for advanced reasoning, coding, and long-running agent tasks.

Coding-focused agentic model built on Kimi K2.6. Strengthens end-to-end task completion across complex software-engineering workflows.

Open-source, native multimodal agentic model. Advances long-horizon coding, coding-driven design, and swarm-based task orchestration.

Built for frontier reasoning, orchestration, and coding agents — up to 5x faster inference and ~30% lower cost on agentic workloads, up to 1M-token context.

Alibaba's most capable Qwen3.5 — a 397B/17B-active MoE with a 262k window (extensible to 1M via YaRN), strong tool calling, and MCP integration.

Z-AI's next-generation flagship for agentic engineering with significantly stronger coding than its predecessor. SOTA on SWE-Bench Pro.

Native omnimodal model with strong agentic capabilities across text, image, video, and audio in a unified architecture.

Google's instruction-tuned Gemma 4 — a capable, efficient open model for general chat, reasoning, and assistant workloads.

Open multimodal model on a hybrid MoE architecture for strong accuracy across image, video, audio, and text — powers always-on sub-agents.