Claim 5M free tokens.Routable. Verifiable. Auditable.
One OpenAI-compatible endpoint for every major model family. Change the Base URL and key, keep your SDK.
- No credit card
- OpenAI-compatible
Many models. One API.
Switch between flagship and lightweight models with one parameter. One key, one account, one quota.
OpenAI Claude Gemini DeepSeek Qwen Zhipu Kimi
Loading service metrics…
Compare prices and context limits. Availability follows your account and group.
Change two settings. Keep the stack.
Model-pinned routing, automatic failover and per-request cost tracing are built in. Your backend, Claude Code, Codex, Cursor and other OpenAI-compatible clients keep working with the client setup you already know.
Before
OPENAI_BASE_URL=https://api.openai.com/v1
OPENAI_API_KEY=sk-proj-••••••••••After — update the URL and key
OPENAI_BASE_URL=https://api.nexrouter.ai/v1
OPENAI_API_KEY=nr-••••••••••Keeps working with the client you use
Python SDKNode SDKCursorClaude CodeCodexLangChainPay for what you use. Top up at 0%.
Top up first, then pay each model’s published rate for the tokens actually used. One balance covers every family and route, and it does not expire.
0%Top-up fee · USD & USDT
- Usage-based
- Pay for tokens
- Each model’s published rate, applied to what you use.
- Prepaid balance
- No expiry
- Top up once. Use your balance across model families.
- Cost visibility
- Every request
- Track model usage and spend in your account.
- Statements list every request at its published rate.
- A 5.5% top-up fee elsewhere? Here, USD and USDT top-ups carry no platform fee.
More routes.
Less interruption.
Pooled capacity, continuous health checks and automatic failover work together to keep requests moving within your routing policy.
Authorized routing. Traffic only on approved lanes.
Route by customer application and reserve dedicated capacity. Requests travel only through authorized provider channels — if no authorized route is available, the request fails rather than falling back to a channel you did not approve.
- Data flow stays inside the authorized boundary.
- Dedicated capacity — critical workloads keep their own capacity.
- Cost by channel or customer — traceable in one place.
The model you pay for. Continuously checked.
Probes send tagged test requests upstream and compare the output with expected fingerprints to detect model downgrades or missing content. When a probe fails, that channel is disabled and cooled down, then re-enabled once a recovery probe passes.
- Continuous checks detect downgrades and missing content on running channels.
- Failing channels are isolated — fewer bad responses reach production.
Channels that fail a probe cool down and return only after a recovery check passes.
5 million free tokens.
Make your first call.
Try DeepSeek, Claude, GPT, Gemini and Qwen. Evaluate routing, reliability and model checks with your own workload before committing to a rollout.
No credit card required. Submit your details; credit is issued after approval.