Fallback strategy

Build robust apps that survive model / channel outages.

Layer 1: Routify auto-fallback (free)

Routify automatically retries within the same model across alternate channels. You don't need to do anything — this is on by default.

Layer 2: Cross-model fallback (your code)

Try a cheaper / different model when premium fails:

from openai import OpenAI
 
client = OpenAI(base_url="https://routify.bytedance.city/v1", api_key="rtf_xxx")
 
MODEL_FALLBACK_CHAIN = [
    "claude-opus-4-7",      # primary
    "claude-sonnet-4-6",    # cheaper alternative
    "kimi-k2.5",            # CN fallback
    "deepseek-v3.2",        # ultra-cheap last resort
]
 
def chat_with_fallback(messages, **kwargs):
    last_err = None
    for model in MODEL_FALLBACK_CHAIN:
        try:
            return client.chat.completions.create(
                model=model, messages=messages, **kwargs
            )
        except Exception as e:
            last_err = e
            continue
    raise last_err

Layer 3: Cross-vendor fallback (advanced)

If Routify itself is down (rare), have a separate OpenAI / Anthropic key as backup. Switch via env var.

Recommended chain by use case

Use case	Chain
Cheap chat	`deepseek-v3.2` → `kimi-k2.5` → `minimax-m2.5`
Premium reasoning	`claude-opus-4-7` → `o1` → `deepseek-r1`
Long context	`qwen3-max` → `kimi-k2.5` → `gemini-3-pro`
Code	`deepseek-v3.2` → `claude-sonnet-4-6` → `qwen3-coder`

Long context FAQ