Leanroute vs Vercel AI Gateway

Native MCP, everything included, one flat bill.

Vercel AI Gateway ships zero token markup, a $5/month free-credit floor, and deep integration with the Vercel AI SDK — a strong default if you already deploy on Vercel. We don't compete on distribution. We compete on what's in the box: native mcp_servers passthrough (they don't forward MCP tool servers), ZDR + provider allowlists + reporting included in the $15 flat sub (Vercel charges per-1k-request for each), APAC + India providers first-class (Vercel is sparse there), and Singapore residency.

Snapshot date: August 2026. Pricing claims re-validated quarterly — check vercel.com/docs/ai-gateway/pricing before depending on a specific number.

TL;DR

Pick Leanroute if (a) you want mcp_servers to just work on any model, (b) you route to APAC or Indian providers, (c) Singapore residency is a hard requirement, or (d) you plan to enable ZDR / allowlists / reporting — where our flat $15 is cheaper than Vercel's per-1k fees above ~75k requests/month. Pick Vercel if (a) you're already on Vercel and the AI SDK integration is worth an ecosystem lock-in, (b) you don't need MCP, (c) you don't need the feature add-ons, or (d) your monthly usage is small enough that $5 free credit + pay-as-you-go tokens beats any flat sub.

Where we differ

DimensionLeanrouteVercel AI Gateway
Token markupZero — provider list priceZero — provider list price
Base plan$15 Starter / $25 Pro flat, BYOK included$5/mo free credit → pay-as-you-go tokens
MCP passthroughNative — add `mcp_servers` to any requestNot supported
Zero Data RetentionIncluded in $15/$25$0.10 per 1k successful requests
Provider allowlistsIncluded$0.10 per 1k successful requests
Custom reporting / analyticsIncluded dashboard + request log$0.075/1k writes, $5/1k queries
Payment-processing fees on top-upsWe absorbPassed through to customer
APAC providers (Qwen, GLM, Doubao, Kimi)First-class, vision includedSparse
India providers (Sarvam, Krutrim)Yes, bothNo
Primary data residencySingaporeUS (Vercel edge)
SDK / wire formatOpenAI-compatible (industry standard)Vercel AI SDK + OpenAI-compatible endpoint
Distribution / ecosystemWord-of-mouth, launched 2026Bundled with Vercel / Next.js — huge default install base
Track recordLaunched mid-2026GA in the Vercel ecosystem since 2024

Rust-tinted cell: Leanroute leads. Muted cell: Vercel leads. No tint: tie or context-dependent.

MCP passthrough

Every LLM gateway on the market routes chat completions. That's the table-stakes feature. As of the snapshot date, Vercel AI Gateway doesn't forward MCP tool servers end-to-end to the model. If you want MCP in your stack, you're wiring it up yourself: a separate MCP client, a separate tool loop, a separate integration to debug when something breaks.

Leanroute takes an mcp_servers field on your existing /v1/chat/completions request — the gateway connects to the MCP server on the model's behalf, handles the tool loop, and stitches the results back into the response. One field. One SDK. One bill. See /docs/mcp for the wire format.

The cost math

Neither of us marks up tokens. Both bill provider list price. The difference is in the feature layer: things like Zero Data Retention, provider allowlists, and custom reporting are per-1k-request add-ons on Vercel, and included in our flat $15/$25 subscription.

Requests / monthLeanroute platform feeVercel platform fee (ZDR + allowlist)You save
10k$15$2Vercel wins by $13
75k$15$15Break-even
500k$15$100$1,020 / yr
5M$25 (Pro)$1,000$11,700 / yr
50M$25 (Pro)$10,000$119,700 / yr

Provider token costs are equal on both sides (both bill list price) and excluded from this table. Payment-processing fees on top-ups are absorbed by us, passed through by Vercel — add ~3.4% on top of every Vercel top-up.

Below ~75k requests/month with ZDR + allowlists enabled, Vercel's per-request add-ons cost less than our flat sub. Above that, our flat line beats their linear one by a widening margin — and if you don't enable the add-ons at all, Vercel is basically free at any volume. The question is whether you need those features, or ever will.

Run your own numbers at /calculator.

Provider catalog

Vercel covers the Western frontier majors (OpenAI, Anthropic, Google, xAI, Groq, Perplexity, DeepSeek). We match that surface plus treat Qwen, GLM, Doubao, Kimi, Sarvam, and Krutrim as first-class with vision routing. If your product ships in Hindi, Tamil, Mandarin, or Bahasa, that's the difference between "works" and "doesn't."

Full catalog at /docs/models.

Where Vercel AI Gateway still wins

  • Distribution. If you already deploy on Vercel and use their AI SDK, their gateway is one env var away. That's a real default effect we can't match on ecosystem alone.
  • Low-volume economics. $5/month free credit + pay-as-you-go tokens with zero add-ons enabled is cheaper than a $15 flat sub for hobby-scale usage. Their pricing is designed to grow with you; ours is designed to plateau.
  • AI SDK ergonomics. The ai package is a nice wrapper for streaming, tool calls, and React server components. We're OpenAI-wire-compatible (works with the AI SDK too), but the deepest integrations are theirs.
  • Track record. GA in the Vercel ecosystem for longer, more third-party examples, larger user pool.

See also: vs OpenRouter · vs Portkey · vs LiteLLM · MCP passthrough docs · Savings calculator · vercel.com/docs/ai-gateway ↗