Model Pool
8 frontier models, each with a specific role. TemuClaude routes to the best model for your question automatically — you never have to choose.
GLM-5.2
The primary orchestrator and fusion aggregator. Highest open-weight IQ (51). Routes queries, aggregates multi-model responses, handles general knowledge at lowest cost.
DeepSeek V4 Pro
The reasoning powerhouse. Best for hard math, coding, and complex logical problems. #1 in Finance on OpenRouter. Used only when needed.
Hy3 Preview
The cheapest model on OpenRouter ($0.063/$0.21). Handles 60% of queries — trivial questions routed here for maximum cost efficiency.
Gemini 3 Flash
Google near-Pro reasoning model. #1 in Legal, #2 in Health. Multimodal (text+image+audio+video+PDF). IQ 50 — nearly matches GLM-5.2.
MiniMax M3
The creative generator and vision specialist. Best GPQA (93%), best instruction following (83%), best hallucination resistance (84%).
MiMo-V2.5
Native omnimodal model from Xiaomi. Processes text, images, and video. Pro-level agentic performance at half the cost.
Claude Sonnet 5
The frontier model for the hardest 2% of queries. IQ 53 — highest in our pool. Used only when fusion + verification + QA all score low.
Nemotron 3 Ultra
The quality gate. Scores answers on 5 rubrics (USVA). 550B/55B hybrid MoE. Completely free — QA and verification costs nothing.