Seventeen models, ten of them Chinese,
one key, one balance, vendor-list pricing.
Chat, image, speech and transcription behind one OpenAI-compatible endpoint. Drop it into Cursor, Cline or OpenClaw by changing base_url. Claude models are currently available through Claude Code only. Every model is priced against what its own vendor charges — most at or below it, with the exact comparison in the live table.
Card or USDT, no Chinese phone
Visa or Mastercard from $5, or USDT on TRC-20 — no Chinese phone number, ID or bank account. The way in when your bank keeps declining the vendor's own checkout. What you pay is what lands in your balance.
One balance, many models
DeepSeek, GLM, Kimi, MiniMax and more on one USDT balance — one bill, no juggling accounts across upstream providers.
OpenAI-compatible
Drop into the OpenAI SDK. Change base_url and key — done. Works with LangChain, AutoGen, CrewAI, OpenClaw, Cline, Continue. Claude models are currently available through Claude Code only.
Shared with GPU rental
One balance covers GPU rentals and API calls. Your first top-up of $10 or more gets a $5 bonus — top up $10, start with $15 to try both.
What every model costs
Chat, image generation, speech synthesis and transcription, all on one balance — the same balance GPU rentals draw from. Each price below is set against what the model's own vendor charges for it — most at or below, and the struck-through figure shows the exact comparison.
Chat & completion
per 1M tokens/v1/chat/completions| Model | Input | Output | Cached input |
|---|---|---|---|
glm-5.3-flash Zhipu GLM | $0.16 | $0.55 | $0.04 |
glm-5.2 Zhipu GLM | $1.26 $1.40 | $3.96 $4.40 | $0.23 |
glm-5.1 Zhipu GLM | $1.26 $1.40 | $3.96 $4.40 | $0.24 |
kimi-k3 Moonshot Kimi | $2.8642 $3.00 | $14.3212 $15.00 | $2.8642 |
kimi-k2.7-code Moonshot Kimi | $0.855 $0.95 | $3.60 $4.00 | $0.855 |
kimi-k2.6 Moonshot Kimi | $0.855 $0.95 | $3.60 $4.00 | $0.4275 |
kimi-k2.5 Moonshot Kimi | $0.855 $0.95 | $3.60 $4.00 | $0.50 |
minimax-m3 MiniMax | $0.2814 $0.30 | $1.1055 $1.20 | $0.1407 |
deepseek-v4-flash DeepSeek | $0.27 $0.30 | $1.08 $1.20 | $0.027 |
deepseek-v4-pro DeepSeek | $1.3065 $1.32 | $3.564 $3.96 | $1.3065 |
gpt-oss-120b OpenAI open-weight | $0.135 $0.15 | $0.54 $0.60 | $0.0675 |
claude-opus-5-5 | $2.00 $4.00 | $10.00 $20.00 | $0.10 |
claude-opus-5 | $2.50 $5.00 | $12.50 $25.00 | $0.25 |
claude-opus-4-8 | $2.50 $5.00 | $12.50 $25.00 | $0.25 |
claude-opus-4-7 | $2.50 $5.00 | $12.50 $25.00 | $0.25 |
claude-opus-4-6 | $2.50 $5.00 | $12.50 $25.00 | $0.25 |
claude-sonnet-5-5 | $1.00 $2.00 | $5.00 $10.00 | $0.10 |
claude-sonnet-5 | $1.00 $2.00 | $5.00 $10.00 | $0.10 |
claude-sonnet-4-6 | $1.50 $3.00 | $7.50 $15.00 | $0.15 |
claude-fable-5-1 | $5.00 $10.00 | $25.00 $50.00 | $0.125 |
claude-fable-5 | $5.00 $10.00 | $25.00 $50.00 | $0.50 |
Image generation
per image/v1/images/generations| Model | Price | Vendor list |
|---|---|---|
flux-1-schnell Black Forest Labs FLUX | $0.0015 | $0.003−50% |
flux-1-dev Black Forest Labs FLUX | $0.0125 | $0.025−50% |
Text to speech
per 1M characters/v1/audio/speech| Model | Price | Vendor list |
|---|---|---|
kokoro-tts Kokoro | $7.50 | $15.00−50% |
chatterbox-tts Chatterbox | $7.50 | $15.00−50% |
Speech to text
per second of audio/v1/audio/transcriptions| Model | Price | Per hour of audio | Vendor list |
|---|---|---|---|
whisper-v3-turbo OpenAI Whisper | $0.00005 | $0.18 | $0.0001−50% |
whisper-v3 OpenAI Whisper | $0.00005 | $0.18 | $0.0001−50% |
All prices in USD, billed from the same balance as GPU rentals. The struck-through figure is what the model's own vendor charges for the same model.
Cached input is charged automatically when the upstream cache fires. Where it equals the input price, the upstream gives us no cache discount, so there is none to pass on.
What a month costs here
Four workload shapes we see in agent and IDE traffic, priced on deepseek-v4-flash against DeepSeek's own list price for the same model.
| Workload | Monthly tokens | Here | DeepSeek list |
|---|---|---|---|
IDE code completion Cursor, Continue.dev, Tabby — inline autocomplete | 18M in + 2M out | $7.02 | $7.80 |
RAG production API LangChain, LlamaIndex, Haystack with 6K-token context | 26M in + 2M out | $9.18 | $10.20 |
Multi-agent code task AutoGen, CrewAI, OpenClaw — agents bouncing per feature | 9M in + 2M out | $4.59 | $5.10 |
Autonomous coding agent Cline, aider, terminal agents reading whole repos | 7M in + 1M out | $2.97 | $3.30 |
Computed from the table above, not typed in — if a price moves, this moves with it. Cache hits make the real bill lower than either column.
Open-source starter · MIT
Deploy a WhatsApp Business autoresponder in 10 minutes
Production-ready Node.js bot — Meta WhatsApp Cloud API + DeepSeek-V4 via cloudgpu. Editable faq.yml for the shop owner, $0 cost on FAQ hits and well under a tenth of a cent per AI reply. One-click deploy to Railway or Fly.io.
Built for small businesses in markets where Stripe-based AI subscriptions keep getting declined. Indonesia, Vietnam, Brazil, Mexico, Nigeria, India, the Philippines — anywhere WhatsApp is the primary customer-comms channel and the going SaaS option starts at $40/mo.
5-minute quick start
curl https://cloudgpu.app/v1/chat/completions \
-H "Content-Type: application/json" \
-H "Authorization: Bearer cgw-sk-..." \
-d '{
"model": "deepseek-v4-flash",
"messages": [{"role": "user", "content": "Hello"}]
}'Common questions
If I have a working Visa, why not just go direct to DeepSeek?
Honest answer: if your card works at platform.deepseek.com and DeepSeek is all you ever need, going direct is fine — we add a hop for no reason. We exist for developers whose cards keep declining (common in SEA, Latin America and Africa), people who can only pay in crypto, and anyone who wants one balance and one key across seventeen models from four vendors instead of four separate accounts. If none of that describes you, go direct.
How does this compare to OpenRouter?
OpenRouter wins on breadth — hundreds of models including the Western ones we do not carry. We are narrower on purpose: Chinese models, priced against each vendor's own list price — most at or below it — with no fee on top of what you send. If you need GPT, OpenRouter has it and we do not. Claude is here at half the official price, currently through Claude Code only.
Is my data sent to mainland China?
Our gateway runs in Hong Kong and forwards each request to its upstream. As of 27 September 2026 every model here is served by DeepInfra in the US, one upstream per model, so there is no automatic failover at the moment. DeepInfra serves each vendor's published weights; for deepseek-v4-flash that is the April open-weights checkpoint, not DeepSeek's newer V4.1 Flash. We do not log prompt or completion content — only token counts, model, latency and error codes for billing. Do not send data that your contracts or laws forbid leaving your jurisdiction. Production-grade DPA available on request.
What payment methods do you accept?
Debit or credit card (Visa / Mastercard, any amount from $5), or USDT on TRC-20 with the same $5 minimum. What you send is credited 1:1 in USD — there is no fee on top; we absorb the network and conversion cost. Requests then deduct from that balance. Bank transfer is also available for business customers.
What about rate limits?
100 requests per minute per API key, 10,000 requests per day per account. These are anti-abuse defaults for the public beta and can be raised on request once you're past the trial.
Will the OpenAI SDK work?
Yes. Use the OpenAI Python or Node.js SDK and override base_url to https://cloudgpu.app/v1. The /v1/chat/completions surface is request-compatible. This covers every model except Claude: Claude models are currently available through Claude Code only.
$5 bonus on a first top-up of $10 or more
Top up $10, start with $15 — roughly 55.6M input tokens on deepseek-v4-flash.
Create your API key