# Claude Code /advisor vs claude-code-router: Which Routing Strategy Wins (May 2026) > **TL;DR:** Anthropic's built-in `/advisor` command picks the right Claude model (Opus 4.5, Sonnet 4.6, Haiku 4.5) for your current turn based on task complexity. claude-code-router is a community proxy that routes to non-Anthropic models too (DeepSeek, Qwen, GLM). Use /advisor for default Anthropic-first workflows; use the router when you need cross-vendor cost arbitrage.

Both tools answer the same question — which model should run this turn? — but they answer it from very different layers. This post walks through what each does, when each wins, and a decision matrix you can paste into your team's playbook.

What is /advisor

/advisor is a first-party slash command shipped with Claude Code (Anthropic's CLI). It inspects the active conversation — file diffs, tool calls, context size, prior thinking depth — and recommends one of the three Anthropic tiers. It runs in-process, no proxy needed.

$ claude
> /advisor

Recommendation: sonnet-4-6
Reason: 12 file edits queued, codebase < 80% context.
Cost vs Opus: 0.33x. Quality delta: ~5% on coding evals.
Apply? [y/N]

It also exposes a non-interactive form so you can pin behavior in ~/.claude/settings.json:

claude --advisor=auto --task "refactor apps/worker/src/scheduler.ts"

What is claude-code-router

claude-code-router (musistudio/claude-code-router on GitHub) is a community-maintained Node proxy. You point Claude Code at it via ANTHROPIC_BASE_URL=http://localhost:3456, and the proxy fans out to whichever provider you've configured: Anthropic, DeepSeek V3, Qwen3-Coder, GLM-4.6, OpenRouter, Ollama, etc.

$ npx @musistudio/claude-code-router start
[router] listening on :3456
[router] default → anthropic/claude-sonnet-4-6
[router] longContext >60k → deepseek/deepseek-v3-0526
[router] background → qwen/qwen3-coder-plus

Routing rules live in ~/.claude-code-router/config.json. The router intercepts the wire protocol, so Claude Code itself never knows it's not talking to Anthropic.

Architectural differences

Layer: /advisor runs inside the CLI. The router runs as a sidecar on the network path.
Vendor scope: /advisor is Anthropic-only. The router is multi-vendor.
Decision input: /advisor sees full conversation state (tool calls, edits, thinking). The router sees only request size, headers, and a model tag.
Failure mode: /advisor degrades gracefully (it just suggests). The router is in the request path — if it crashes, every turn fails.
Cost telemetry: /advisor reports against your Anthropic spend. The router needs you to wire your own logging (Helicone, Langfuse, etc.).

When /advisor wins

Pick /advisor when you want zero ops overhead, deterministic billing, and the latest Anthropic features (computer use, code execution tool, MCP cache). It's also the only routing layer that can use the conversation's actual content — claude-code-router can't tell whether your turn is a one-line typo fix or a 600-line refactor.

If your team standardized on Claude Opus 4.5 / Sonnet 4.6 / Haiku 4.5, /advisor typically saves 30-50% on monthly spend by demoting trivial turns to Haiku. See the Claude Code /advisor command deep-dive for the full rule table.

When router wins

Pick claude-code-router when:

You want to use DeepSeek V3 or Qwen3-Coder for bulk codegen and pay $0.27 / Mtok input instead of $3.
You're on the Pro plan ($20/mo) and want to spill overflow to OpenRouter.
You need air-gapped inference via Ollama for compliance.
You want different models for longContext, background, and think phases — the router exposes these hooks; /advisor does not.

Decision matrix

Use /advisor if:
  - 100% Anthropic, want lowest config friction
  - Need conversation-aware model picking
  - Want first-party support & SLA

Use claude-code-router if:
  - Mixing Anthropic + DeepSeek/Qwen/GLM
  - Need cost arbitrage on long-context turns
  - Want air-gapped or self-hosted fallback
  - OK running a sidecar process

Use BOTH if:
  - Run /advisor for Anthropic-tier picking,
    then router for vendor-tier fallback when
    rate limits hit. Stack: CLI → /advisor →
    router → provider.

In practice, most indie hackers we surveyed in April 2026 ran /advisor alone. Teams burning >$2k/month on Claude Code reached for the router to capture DeepSeek's price floor on bulk refactors.

FAQ

Can I use /advisor and claude-code-router together?

Yes. Set ANTHROPIC_BASE_URL to the router and let /advisor emit a model tag the router then maps. Just confirm your router config has rules for opus-4-5, sonnet-4-6, and haiku-4-5.

Does /advisor work with non-Anthropic models?

No. As of May 2026, /advisor only ranks across opus-4-5, sonnet-4-6, and haiku-4-5. For cross-vendor routing you need the community router or a custom MCP gateway.

Which is cheaper, /advisor or claude-code-router?

For Anthropic-only stacks, /advisor is usually 5-15% cheaper because it picks Haiku 4.5 more aggressively than humans do. For mixed stacks, the router wins by routing 60-80% of turns to DeepSeek V3 at one-tenth the price.

Will /advisor replace claude-code-router?

Unlikely in 2026. Anthropic has no roadmap commitment to non-Anthropic routing inside the CLI. The router fills a real gap and the community is shipping faster than Anthropic on multi-vendor features.

#claude-code-router

Claude Code /advisor vs claude-code-router: Which Routing Strategy Wins (May 2026)

#claude-code-router

Claude Code /advisor vs claude-code-router: Which Routing Strategy Wins (May 2026)