Leanroute vs Vercel AI Gateway
Vercel AI Gateway alternative: Leanroute compared.
Vercel AI Gateway ships zero token markup, a $5/month free-credit floor, and deep integration with the Vercel AI SDK. It is a strong default if you already deploy on Vercel. We don't compete on distribution. We compete on what's in the box: team-wide ZDR, provider allowlists and reporting included in a flat $15/$25 subscription (Vercel bills each per 1k requests), guardrails enforced at the gateway (Vercel has none natively), MCP tools run at the gateway on any model, APAC + India providers first-class, and Singapore hosting.
Snapshot date: October 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
| Dimension | Leanroute | Vercel AI Gateway |
|---|---|---|
| Token markup | Zero — provider list price | Zero — provider list price |
| Base plan | $15 Starter / $25 Pro flat, BYOK included | $5/mo free credit → pay-as-you-go tokens |
| MCP tool calling | At the gateway: add `mcp_servers` to any chat-completions request and the gateway runs the tool loop on any model. Per-org server allowlist. | In your app: the AI SDK's MCP client discovers and runs tools; AI Gateway routes the model calls |
| Guardrails | 7 curated rules (PII, secrets, prompt injection, moderation, data exfiltration, profanity, language gate), enforced at the gateway | No native guardrails; moderation models can be called through it |
| Zero Data Retention | Included in $15/$25 | Free per request; $0.10 per 1k requests to enforce team-wide |
| Provider allowlists | Included | $0.10 per 1k successful requests |
| Custom reporting / analytics | Included dashboard + request log | $0.075/1k writes, $5/1k queries |
| APAC + India providers | First-class: Qwen, GLM, Doubao, Kimi, Sarvam, Krutrim (vision included) | Check their model list for your mix |
| Primary data residency | Singapore | US (Vercel edge) |
| SDK / wire format | OpenAI-compatible (industry standard) | Vercel AI SDK + OpenAI-compatible endpoint |
| Distribution / ecosystem | Word-of-mouth, launched 2026 | Bundled with Vercel / Next.js — huge default install base |
| Track record | Launched mid-2026 | GA in the Vercel ecosystem since 2024 |
Rust-tinted cell: Leanroute leads. Muted cell: Vercel leads. No tint: tie or context-dependent.
Where MCP runs
Vercel keeps MCP out of the gateway on purpose. In their stack, the AI SDK's MCP client discovers tools and runs the tool loop inside your application, and AI Gateway only routes the model calls. If your app is built on the AI SDK, that is a clean design. The cost is that every service that wants MCP needs the SDK and its own tool loop, and the gateway never sees the tool calls.
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 / month | Leanroute platform fee | Vercel platform fee (ZDR + allowlist) | You save |
|---|---|---|---|
| 10k | $15 | $2 | Vercel wins by $13 |
| 75k | $15 | $15 | Break-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
aipackage 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 docs · Savings calculator · vercel.com/docs/ai-gateway ↗