Claude Code

Run Claude Code with Routify-hosted Anthropic models, or substitute with DeepSeek / Kimi for 5-10× cost reduction.

Setup (Anthropic-direct, mirror)

export ANTHROPIC_BASE_URL=https://routify.bytedance.city/v1/anthropic
export ANTHROPIC_API_KEY=rtf_xxx
 
claude

Routify's /v1/anthropic endpoint speaks the Anthropic Messages protocol natively, so Claude Code works unchanged.

Setup (DeepSeek substitution)

If you want to use DeepSeek instead of Claude under the hood:

export ANTHROPIC_BASE_URL=https://routify.bytedance.city/v1/anthropic
export ANTHROPIC_API_KEY=rtf_xxx
export ANTHROPIC_MODEL=deepseek-v3.2  # gets translated server-side
 
claude

Routify will automatically translate the Anthropic protocol to OpenAI protocol, call DeepSeek, and translate the response back.

⚠️

DeepSeek doesn't have the same tool-calling semantics as Claude — agentic flows like TaskTool may behave differently. Keep claude-sonnet-4-6 for production agents.

Per-conversation model

Use /model slash command in Claude Code to switch models mid-session:

/model deepseek-v3.2
/model claude-opus-4-7
/model kimi-k2.5

Cost comparison (typical CC session)

Model	Cost per 100k tokens	vs Claude Opus official
`claude-opus-4-7` (Routify)	$7.50	1.0× (no markup)
`claude-sonnet-4-6` (Routify)	$1.50	0.2×
`deepseek-v3.2` (Routify)	$0.04	0.005× (200× cheaper)
`kimi-k2.5` (Routify)	$0.31	0.04× (24× cheaper)

Cursor Cline