DeepSeek V4.1 Flash is a multimodal MoE model (763B total, 16B active) with a 1M-token context and an FP4 KV cache built for extreme long-context efficiency.

DeepSeek V4 Flash is a MoE model (284B total, 13B active) with a 164K-token context, tuned for low-latency, cost-efficient inference.

DeepSeek V4 Pro is a MoE model (1.6T total, 49B active) with a 164K-token context and hybrid compressed sparse attention.

Gemma 4 31B is a 31B dense multimodal model with a 262K-token context, configurable reasoning, and 140+ language support.

GLM 5.2 is an open-weight MoE model (753B total, 40B active) with a 1M-token context, built for coding and long-horizon agent workflows.

GLM 5.3 is an open-weight MoE model (753B total, 40B active) with a 1M-token context, Z.ai's most capable model for complex coding and long-horizon agents.

GLM 5.3 Flash is a natively multimodal MoE model (320B total, 18B active) with a 1M-token context and hybrid linear-sparse attention for efficient inference.

Kimi K2.6 is a native multimodal MoE model (1T total, 32B active) with agent swarms scaling to 300 sub-agents and 4,000 coordinated steps.

Kimi K2.7 Code is a coding-focused MoE model (1T total, 32B active) that cuts reasoning tokens 30% while improving agentic coding accuracy.

Kimi K3 is a MoE model (2.8T total, 104B active) with a 1M-token context, native visual understanding, and always-on reasoning.

MiniMax H3 is an omni-modal video model generating 15-second 2K clips with native stereo audio from text, image, video, or audio input.

MiniMax M3 is a natively multimodal MoE model (427B total, 26B active) with a 1M-token context and frontier coding performance.

Qwen3.5 is a multimodal MoE model (397B total, 17B active) with a 262K-token context and strong reasoning, coding, and vision-language performance.

Qwen3.8 is a sparse MoE model (2.4T total, 95B active) and the open-weight flagship of the Qwen family, with a 262K-token context.