China Frontier
DeepSeek · GLM · Kimi · Qwen — exceptional value and long context
DeepSeek
CN · Flagship
DeepSeek V4 Pro
SWE-bench 80.6%, roughly 1/10 the cost of Opus
1M contextCoding leaderOpen weights
1M context$0.2800 in · $2.4800 out / 1M
Z.AI
CN · Flagship
GLM-5.2
MIT open source, enterprise self-hosting and sovereign deployment
MIT licenseSelf-hostedEnterprise ready
200K context$0.6300 in · $2.5000 out / 1M
Moonshot
CN · Flagship
Kimi K2.6
Native swarm sampling for complex multi-step workflows
Agent swarmMulti-step tasks256K context
256K context$0.3000 in · $0.9000 out / 1M
Qwen
CN · Flagship
Qwen 3.6 Plus
Broadest product line, bilingual long context and tool use
1M contextBilingualTool use
1M context$0.1500 in · $0.4500 out / 1M
MiniMax
CN · Flagship
MiMo V2.5 Pro
Lightweight and efficient low-latency inference
Low latencyEfficientCost effective
128K context$0.1400 in · $0.4200 out / 1M
One line to switch
Switch models with one string
// China frontier
{ "model": "deepseek/deepseek-v4-pro" }
{ "model": "zhipu/glm-5-2" }
{ "model": "moonshot/kimi-k2-6" }
{ "model": "alibaba/qwen-3-6-plus" }
// US frontier
{ "model": "openai/gpt-5.5" }
{ "model": "claude-opus-4-8" }
{ "model": "gemini-3-1-pro" }* Claude Fable 5 is not publicly available due to export controls. ModGate routes Claude Opus 4.8 as the Anthropic flagship alternative.