OpenAI-compatible API. One model. A dedicated virtual server per account " it's yours alone, no other tenants.
Dashboard → API keys → Create. Keys label your tasks (observability). They do not cap concurrency — concurrency is your product's max-num-seqs.
base_url: https://api.notoken.cloud/{instance_id}/v1
api_key: tnt-...
model: qwen3.8-27b
Your instance_id is unique to your server and appears in the dashboard the moment it goes live. The API key authenticates your account; the instance id routes to your server.
Works with: Cursor, Claude Code (via claude-code-router), Cline, Roo Code, Aider, OpenCode, Windsurf, Zed, Factory Droid, LiteLLM, Portkey, Braintrust, Vercel AI Gateway " anything that speaks OpenAI chat completions.
curl https://api.notoken.cloud/{instance_id}/v1/chat/completions \
-H "Authorization: Bearer $TNT_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "qwen3.8-27b",
"stream": true,
"messages": [{"role": "user", "content": "Refactor this module: ..."}]
}'
Streaming is SSE. Long prefills (big contexts) may pause before the first token " keepalives are injected so proxies (including Cloudflare) don't drop the connection. Hard cap: 10 min per request.
Native: 262,144 tokens. Extended: up to 1,000,000 (YaRN RoPE scaling, factor 4.0 " Qwen's recommended method). Per-product practical limits are on the pricing page. For agentic work, Qwen recommends reserving up to 262K for reasoning + 131K for the final response inside the 1M window.
Prepaid credits (1 credit = $0.01). Your server accrues its product rate per second while running. Auto-stop: your configured idle time (default 15 min), or credits = 0 (60s grace so in-flight generations finish). Top up from the dashboard (Stripe) " the server restarts on demand. Per-second billing, no egress fees, no minimums beyond the $10 top-up.
OpenAI-compatible API · up to 1M context · auto-stop on your rules · AI support · per-second billing with prepaid credits.
| Status | Meaning |
|---|---|
| 401 | Missing/invalid API key |
| 409 | No active server (launch one) or server already exists |
| 402 | Insufficient credits |
| 503 | Server not ready yet " retry in a few seconds |
| 502 | Server unreachable (it may be restarting) |
Qwen3.8 is a beast — scoring 52 on the intelligence index (alongside GLM-5.2), it is configured on noTOKEN.cloud with xhigh reasoning and all the tweaks expected for long-horizon tasks to evolve continuously. That means your server is expected to work autonomously, on its own, after your prompt.
If you give it detailed instructions, it will Plan, Build, Review and Document the entire task for you — sometimes working for hours (a single Blackwell user on our servers has an average ~110 tok/s generation speed — no token limits, as our name says).
If you want the model to keep tasks short, instead of changing the API reasoning flags per call, you can simply add to your prompt just do this and then stop — a simple instruction, just like you would give to a person.
1 million tokens in the context window is wonderful for coding, but it is too big for some tasks. For instance, if you want your Qwen to answer customer support messages, anything greater than 90K tokens can feed the model years of conversation that might not be useful for such a short task. Note also that prompt processing time is a quadratic feature: under the Ada architecture (Starter plan) it can take several minutes to process, while under Blackwell plans it is much faster.
As expected, the fastest prefill and generation occur on the Ultra plan — that is where the GPU memory bandwidth is the fastest. For more information, check our Benchmark section on this website.