DeepSeek: DeepSeek V4 Flash 0423 in Claude Code
Run deepseek/deepseek-v4-flash as the model behind Claude Code through Redline, with a hard spend cap on the key and no prompt retention.
V4 Flash is a cheap, very-long-context reasoning model, verified with Claude Code in testing. Its reasoning shows as thinking blocks within a turn but does not carry between turns through the gateway. A million-token context for the price makes it good for large repos.
| Model id | deepseek/deepseek-v4-flash |
|---|---|
| Context | 1M tokens |
| Input price | $0.102 / million tokens |
| Output price | $0.204 / million tokens |
| Tool calling | Supported |
Set it up
Point Claude Code's base URL at Redline (no /v1), set the auth token to your key, name this model, and set the context length to its real value of 1,048,576:
# Put these in your shell profile, then run: claude
export ANTHROPIC_BASE_URL=https://redline-gateway.pages.dev
export ANTHROPIC_AUTH_TOKEN=rl-...
export ANTHROPIC_MODEL=deepseek/deepseek-v4-flash
export ANTHROPIC_DEFAULT_HAIKU_MODEL=qwen/qwen3-30b-a3b-instruct-2507
export CLAUDE_CODE_MAX_CONTEXT_TOKENS=1048576
Or put the same values under env in ~/.claude/settings.json:
// ~/.claude/settings.json
{
"env": {
"ANTHROPIC_BASE_URL": "https://redline-gateway.pages.dev",
"ANTHROPIC_AUTH_TOKEN": "rl-...",
"ANTHROPIC_MODEL": "deepseek/deepseek-v4-flash",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "qwen/qwen3-30b-a3b-instruct-2507",
"CLAUDE_CODE_MAX_CONTEXT_TOKENS": "1048576"
}
}
ANTHROPIC_DEFAULT_HAIKU_MODEL is the cheap background model, set to qwen/qwen3-30b-a3b-instruct-2507 above. CLAUDE_CODE_MAX_CONTEXT_TOKENS is set because Claude Code assumes 200K for an id it does not recognise. Use ANTHROPIC_AUTH_TOKEN and leave ANTHROPIC_API_KEY unset. Full setup and every caveat is in the Claude Code guide.
FAQ
Can DeepSeek: DeepSeek V4 Flash 0423 edit files and run commands in Claude Code?
Yes. It supports tool calling, which is what Claude Code needs to use its Read, Edit, Write and Bash tools. A model without tool calling could only chat.
How much does it cost?
$0.102 per million input tokens and $0.204 per million output tokens through Redline, margin already included. Put a spend cap on the key so an agent loop cannot overrun it.
Is this an official Anthropic setup?
No. Anthropic does not officially support routing Claude Code to non-Claude models through a gateway. It works because Redline translates the Messages API, but it is not a configuration Anthropic backs.