Premium models with credits
Your usual model stays dollar-billed. If you pass one of the model names below instead, the request pays with credits — a second wallet you top up in packs. Model names identify performance-and-price tiers served on our infrastructure: top-tier names think at full depth, light names answer fast and cheap.
Using a model
Same endpoint, same fields. Put the model name in model:
curl https://api.aiprovider.org/v1/chat/completions \
-H "Authorization: Bearer $AI_PROVIDER_KEY" \
-H "Content-Type: application/json" \
-d '{
"model": "claude-sonnet-5.5",
"messages": [{"role": "user", "content": "Summarise this in two sentences."}]
}'The reply comes back with the model name you asked for, and usage counts the tokens it really used. Every model takes the same requests as aiprovider-1 — streaming, tool calling, JSON mode, reasoning all work the same way.
Rates
Charged per token at these rates, rounded up, minimum 1 credit per request. Buying credits: 500 for $5, 1,000 for $10, 2,600 for $25, 5,300 for $50, or any whole amount from $5 to $1,000.
| Model | Maker | Credits / M sent | Credits / M back | Typical reply |
|---|---|---|---|---|
claude-fable-5.1 | Anthropic | 3,500 | 17,500 | ~20 |
gpt-6-astra | OpenAI | 3,500 | 17,500 | ~20 |
claude-opus-5.5 | Anthropic | 1,400 | 7,000 | ~8 |
kimi-k3 | Moonshot | 56 | 7,000 | ~6 |
gpt-6.1-sol | OpenAI | 700 | 3,500 | ~4 |
gemini-4-argon | 700 | 3,500 | ~4 | |
claude-sonnet-5.5 | Anthropic | 700 | 3,500 | ~4 |
kimi-k3.1-pro | Moonshot | 700 | 3,500 | ~4 |
glm-5.3-prime | Z.ai | 980 | 3,080 | ~4 |
grok-4.7 | xAI | 700 | 2,100 | ~3 |
gpt-6.1-terra | OpenAI | 350 | 1,750 | ~2 |
muse-spark-1.3 | Meta | 438 | 1,488 | ~2 |
gemini-3.8-flash | 291 | 1,446 | ~2 | |
step-5-preview | StepFun | 350 | 945 | ~2 |
hy4-preview | Tencent | 291 | 875 | ~2 |
nemotron-3-ultra | NVIDIA | 210 | 840 | ~1 |
deepseek-v4.1-flash-1017 | DeepSeek | 105 | 420 | ~1 |
deepseek-v4-flash-0731 | DeepSeek | 154 | 462 | ~1 |
glm-5.3 | Z.ai | 25 | 490 | ~1 |
minimax-m3.1-flash | MiniMax | 53 | 210 | ~1 |
hy3 | Tencent | 46 | 186 | ~1 |
glm-5.3-flash | Z.ai | 53 | 175 | ~1 |
gpt-6-luna | OpenAI | 35 | 175 | ~1 |
claude-haiku-5.5 | Anthropic | 35 | 175 | ~1 |
deepseek-v4.1-flash | DeepSeek | 49 | 147 | ~1 |
mimo-v2.6-flash | Xiaomi | 49 | 98 | ~1 |
deepseek-v4-flash | DeepSeek | 49 | 98 | ~1 |
muse-spark-1.3-contributor | Meta | 35 | 70 | ~1 |
How credits work
- Credits and credit are separate wallets: model requests never spend your dollar credit, and normal requests never spend credits.
- A running request can take the credit balance slightly below zero; it comes out of your next pack.
- Out of credits, requests stop with a clear 402 until you buy a pack. No surprise bills.
- Credits never expire. Credit purchases are final.
- Turn on credit auto top-up on the Billing page and we buy the pack you chose on your saved card when the balance runs down, at most once every 10 minutes.
- Every credit request shows in your dashboard Activity page with its model and credit cost.
Buy credits any time on the Billing page. The public rate table is on the pricing page.