OpenAI-compatible proxy API

Every model,
one key.

The full model pool behind a single OpenAI-compatible endpoint. Swap your base URL and your code stays byte-for-byte the same.

Start building → Check a key
# one line changes — the rest of your code doesn't
export OPENAI_BASE_URL="https://openai.mrvinci.app"
export OPENAI_API_KEY="sk_mv_…"
The lineup

The whole model pool

Pass any id below as the model; usage is metered per key.

gpt-5.6-solvision · flagship

Top of the 5.6 line — deepest reasoning in the pool.

high effort
gpt-5.6-terravision

Balanced 5.6 tier — flagship quality at mid cost.

high effort
gpt-5.6-lunavision · fast

Fast, affordable 5.6 — a great high-volume default.

low effort
gpt-5.5vision

Frontier general reasoning. Accepts image input.

high effort
gpt-5.4vision

Strong multimodal all-rounder.

high effort
gpt-5.4-minivision · fast

Fast, cheap, multimodal.

low effort
gpt-5.3-codexcoding

Agentic coding specialist, latest codex line.

high effort
gpt-5.1-codexcoding

Codex tuned for tool-use and long edits.

medium effort
gpt-5.1vision

Balanced multimodal reasoning.

medium effort
gpt-5-minivision · low cost

Cheapest multimodal GPT, high throughput.

low effort
DeepSeek-V4-Pro1M context

1M-token context, top-tier agentic coding.

high effort
DeepSeek-V4-Flashlow cost

High throughput, lowest cost in the pool.

low effort
Kimi-K2.7-Codevision · code

Vision + code specialist.

high effort
Kimi-K2.6efficient

Capable reasoning generalist at low cost.

medium effort
grok-4.3fast

Fast, current-events aware.

medium effort
Mistral-Large-3efficient

Efficient European model.

medium effort

Image models

How it flows

One key, the whole lineup

Your agent talks to one endpoint. Every request is metered against your key and the answer streams straight back — any model in the lineup, same key.

sk_mv_…
OpenAI API gateway
DeepSeekgpt-5.xKimi
Why MrVinci

Built for teams that need control

Multi-tenant keys

Mint a key per team or per app. Revoke or top up any one without touching the rest.

Per-key budgets & limits

Prepaid credit per key that debits at model pricing and stops when it's spent. Keys can carry an expiry date.

Streaming, native

Server-sent events with no added latency — text, tools stream exactly as your SDK expects.

Vision, handled

Send screenshots and images exactly as the API specifies — vision requests just work.

SDK & IDE compatible

Works with the official SDKs and with Claude Code, Claude Desktop, Cursor, and Cline — only the base URL changes.

Live usage

Every key has a status page: spend or window consumption, per-model breakdown, and request counts.

Setup

Live in about a minute

Get a key

Ask an admin for an sk_mv_ key, or mint one in the MrVinci admin panel.

Point your client at us

Run npx mrvinciapi --openai and paste the key, or set OPENAI_BASE_URL yourself.

Ship

Start a new session — your traffic now flows through the gateway. No code changes.

Read the docs →