Guides
Choosing a model

Choosing a model

Decision tree:

1. Is it a coding task?

→ deepseek-v3.2 (default) → fall back to claude-sonnet-4-6 for hard tasks

2. Need long context (> 64k tokens)?

ContextModel
128kkimi-k2.5 ($0.60/M), glm-4.6 ($0.50/M)
256kqwen3-max ($0.60/M)
200kclaude-opus-4-7 ($15/M) — premium
2Mgemini-3-pro ($1.25/M)

3. Need step-by-step reasoning?

→ deepseek-r1 ($0.55/M, cheap) or o1 ($15/M, premium)

4. Need tool / function calling?

Best in class:

  • kimi-k2.5 — solid + cheap
  • glm-4.6 — China's most-tuned for agents
  • claude-sonnet-4-6 — premium

Avoid:

  • deepseek-v3.2 — tool calling works but flakier than Kimi/GLM
  • o1 — doesn't support tools yet

5. Need images / multimodal?

  • doubao-1.6-pro — China's multimodal champ
  • gpt-4o — global standard
  • gemini-3-pro — best for image+long-context combo

6. Need video generation?

→ kling-1.6 — currently only video model on Routify

7. Just want cheap?

→ deepseek-v3.2 ($0.14/M input). It's the default for a reason.

8. Want Routify to pick automatically?

client.chat.completions.create(model="routify-auto", ...)

We pick the lowest-cost healthy channel for your task class. Specify class:

extra_body={"routify": {"task": "code"}}  # chat | code | reasoning | longContext | vision
RaycastSmart routing