Top-up bonus
Extra credit on every top-up
Applies to top-ups of $5 or more.
Bonus credit is added to your balance on top of what you pay and can be spent the same way.
Text · Moonshot / KimiAdded today
$0.465 / 1M input
Access DeepSeek, Qwen, Kimi, GPT, Claude, Grok and more through a single OpenAI-compatible endpoint. You pay per token you actually use — one balance, no subscription, $1.50 to start.
Built for AI apps and the tools developers already run — Claude Code, Cursor, Cline, Codex and anything else that speaks the OpenAI or Anthropic API.
Base URL: https://jzstoken.com/api/v1
Two key types, and they are mutually exclusive: a Standard key covers every model except jzs-max-3.0, which needs its own dedicated key. Your balance is shared across all your keys either way.
10 of 66 models we currently sell — picked to cover the common jobs, not by sales rank. Every model below has been called successfully through this platform. Prices are in USD per 1M tokens and come straight from the billing table.
| Model | Best for | Input / 1M | Output / 1M | Cache read / 1M |
|---|---|---|---|---|
| gpt-5.6-terra OpenAI | balanced general-purpose flagship | $0.2325 | $1.39 | $0.0232 |
| gpt-5.5 OpenAI | reliable everyday chat and coding | $0.3875 | $2.32 | $0.0387 |
| jzs-max-3.0 JZS Token dedicated key | our flagship, dedicated capacity | $0.0620 | $0.3100 | $0.0062 |
| claude-opus-5 Anthropic | hardest reasoning and long documents | $1.16 | $5.81 | $0.1162 |
| claude-haiku-4-5-20251001 Anthropic | cheapest Claude, high volume | $0.0775 | $0.3875 | $0.0077 |
| grok-4.6 xAI / Grok | reasoning with a different bias | $0.3875 | $1.94 | $0.4844 |
| LongCat-2.0 Meituan / LongCat | lowest-cost general chat | $0.0775 | $0.3100 | $0.0077 |
| mimo-v2.5 Xiaomi / MiMo | budget long-context work | $0.1550 | $0.7750 | $0.0155 |
| MiniMax-M3 MiniMax | cheap high-throughput batch jobs | $0.0775 | $0.3875 | $0.0077 |
| gpt-image-2 OpenAI | image generation | $0.0116per image | — | — |
Each row links to full specs: cache-write rate, peak-hour rules, context window, image support and the exact request example. Cache-read rates assume the upstream provider reports cached tokens — how billing works. Live USD→CNY rate 6.71, prices last updated 2026-10-08.
The same path for every client. If you would rather see it all at once, the Quick Start has the copy-paste version.
Step 1
Register, then create a key in API Keys. Pick Standard unless jzs-max-3.0 is the model you want — that one needs its own key.
Step 2
Use https://jzstoken.com/api/v1 as the base URL and your key as the API key. Client-by-client steps are in the setup guides, or run the curl example.
Step 3
Top-ups start at $1.50 and credits do not expire. Token counts and cost per call show up under Usage, and you can look up any key's balance on the balance page.
Training data cutoffs, context limits and per-model quirks are on each model's page — open any row in the model catalog.
Any client that accepts a custom OpenAI-compatible base URL can point at this endpoint. These are step-by-step guides for the clients our customers ask about most. We have not individually re-tested every client or every optional feature (streaming, tool calling, vision), so treat each guide's scope note as part of the instructions rather than reading this list as a compatibility claim.
Requests are handled in memory; we keep billing metadata only (the Studio keeps your prompts for your history). We forward content to the upstream provider of the model you pick, and their policy applies to that leg — we do not claim to control it. Privacy policy
Reach us on WhatsApp or by email. Hours: Daily, 09:00–22:00. All contact options
Charges are per token from a prepaid balance that does not expire. Failed payments, duplicate charges and delivery problems are investigated and fixed; refund policy and terms.
JZS Token is an API reseller: we route your requests to the providers of the models you choose. We are not affiliated with, endorsed by, or an official partner of those providers — model names belong to their owners. We do not publish an uptime figure because we do not operate a status page; if a call fails, the error body says whether it came from us or from upstream.
