LLM API pricing, billed per token
Each model has its own input and output token rates. Top up $1 for $10 in credit; the request log records the exact credit deducted per call.
$1 of sign-up credit ≈ 10M input tokens on openai/gpt-6-luna, the lowest-priced model in the catalog at $0.10 per 1M tokens in credit. Every $1 you top up becomes $10 in credit ≈ 100M input tokens on that same model.
| You pay | Credit received |
|---|---|
| $5 | $50 |
| $10 | $100 |
| $50 | $500 |
| $100 | $1,000 |
Not a plan tier — there is no amount threshold; any top-up converts at the same rate.
- Billing
- Credit is deducted per token of usage. No calls, no charges.
- Minimum spend
- None. An idle account costs nothing either.
- New users
- $1 credit on sign-up, no card required.
- Credit exhausted
- The API returns 403 (insufficient_user_quota). Top up at https://y-api.bestvirtualgoods.com/app/billing — it recovers immediately, and keys stay valid.
What a dollar buys
Every model, priced in what you actually pay
The rates on the models page are in USD credit. Divide by the 1:10 top-up rate and you get the cash column below — the dollars that leave your card per 1M input tokens. The third column is the same figure read the other way round: how many input tokens one dollar buys.
| Model ID | You pay / 1M input | $1 buys | vs vendor official |
|---|---|---|---|
| Vendor: DeepSeek | |||
| deepseek/deepseek-v4-flash | Free — no credit deducted | $0.221,2,3off-peak rate; a cheaper cache-hit tier also exists; historical official price for this version; current rolling-model pricing may differ | |
| deepseek/deepseek-v4-pro | $0.05 | 20M | 13.2×$0.661,2off-peak rate; a cheaper cache-hit tier also exists |
| deepseek/deepseek-v4.1-flash | $0.02 | 50M | 7.5×$0.151,2off-peak rate; a cheaper cache-hit tier also exists |
| deepseek/deepseek-v4-flash-0731 | $0.015 | 66.7M | 14.7×$0.221,2,3off-peak rate; a cheaper cache-hit tier also exists; historical official price for this version; current rolling-model pricing may differ |
| Vendor: Qwen | |||
| qwen/qwen3.8-max | $0.20 | 5M | 8.25×$1.655cheapest of its six regional price tables |
| qwen/qwen3.8-flash | $0.02 | 50M | 5.65×$0.1135cheapest of its six regional price tables |
| Vendor: MiniMax | |||
| minimax/minimax-m2.7 | Free — no credit deducted | $0.302a cheaper cache-hit tier also exists | |
| minimax/minimax-m3 | $0.03 | 33.3M | 10×$0.302,6a cheaper cache-hit tier also exists; cheapest input-length tier; a longer prompt costs more per token |
| Vendor: Z.ai | |||
| z-ai/glm-5.1 | $0.14 | 7.14M | 10×$1.402a cheaper cache-hit tier also exists |
| z-ai/glm-5.2 | $0.14 | 7.14M | 10×$1.402a cheaper cache-hit tier also exists |
| z-ai/glm-5.3 | $0.14 | 7.14M | 10×$1.402a cheaper cache-hit tier also exists |
| z-ai/glm-5.3-flash | $0.015 | 66.7M | 10×$0.152a cheaper cache-hit tier also exists |
| Vendor: Moonshot AI | |||
| moonshotai/kimi-k3 | $0.30 | 3.33M | 10×$3.002a cheaper cache-hit tier also exists |
| moonshotai/kimi-k2.6 | $0.095 | 10.5M | 10×$0.952a cheaper cache-hit tier also exists |
| Vendor: Anthropic | |||
| anthropic/claude-opus-4.6 | $0.50 | 2M | 10×$5.002a cheaper cache-hit tier also exists |
| anthropic/claude-opus-4.7 | $0.50 | 2M | 10×$5.002a cheaper cache-hit tier also exists |
| anthropic/claude-opus-5 | $0.50 | 2M | 10×$5.002a cheaper cache-hit tier also exists |
| anthropic/claude-sonnet-5 | $0.20 | 5M | 10×$2.002a cheaper cache-hit tier also exists |
| anthropic/claude-haiku-4.5 | $0.10 | 10M | 10×$1.002a cheaper cache-hit tier also exists |
| anthropic/claude-haiku-5.5 | $0.015 | 66.7M | 6.67×$0.102,6a cheaper cache-hit tier also exists; cheapest input-length tier; a longer prompt costs more per token |
| Vendor: Google | |||
| google/gemini-3.1-pro-preview | $0.20 | 5M | 10×$2.002,6a cheaper cache-hit tier also exists; cheapest input-length tier; a longer prompt costs more per token |
| google/gemini-3.5-flash | $0.20 | 5M | Not on the vendor’s rate card |
| google/gemini-3.7-flash | $0.15 | 6.67M | Not on the vendor’s rate card |
| google/gemini-3.8-flash | $0.15 | 6.67M | 5×$0.752,7a cheaper cache-hit tier also exists; limited-time promotional rate; the list price is higher |
| Vendor: OpenAI | |||
| openai/gpt-5.6-sol | $0.50 | 2M | 8×$4.002,4,6,7a cheaper cache-hit tier also exists; standard tier; priority and flex tiers differ; cheapest input-length tier; a longer prompt costs more per token; limited-time promotional rate; the list price is higher |
| openai/gpt-5.6-luna | $0.03 | 33.3M | 6.67×$0.202,4,6a cheaper cache-hit tier also exists; standard tier; priority and flex tiers differ; cheapest input-length tier; a longer prompt costs more per token |
| openai/gpt-6-luna | $0.01 | 100M | 10×$0.102,4,6a cheaper cache-hit tier also exists; standard tier; priority and flex tiers differ; cheapest input-length tier; a longer prompt costs more per token |
| Vendor: xAI | |||
| x-ai/grok-4.6 | $0.40 | 2.5M | 5×$2.002,6a cheaper cache-hit tier also exists; cheapest input-length tier; a longer prompt costs more per token |
| x-ai/grok-4.7 | $0.20 | 5M | 10×$2.002,6a cheaper cache-hit tier also exists; cheapest input-length tier; a longer prompt costs more per token |
| Vendor: Other | |||
| tencent/hy3 | Free — no credit deducted | $0.20 | |
| xiaomi/mimo-v2.5 | Free — no credit deducted | $0.142a cheaper cache-hit tier also exists | |
| tencent/hy4-preview | $0.10 | 10M | 12×$1.20 |
| xiaomi/mimo-v2.6-pro | $0.05 | 20M | 8.7×$0.4352a cheaper cache-hit tier also exists |
| stepfun/step-3.7-flash | $0.02 | 50M | 11×$0.22 |
| xiaomi/mimo-v2.6-flash | $0.018 | 55.6M | 8.89×$0.16 |
Vendor official prices verified 2026-08-31–2026-10-11. Sources are listed on the price comparison page.
- 1off-peak rate
- 2a cheaper cache-hit tier also exists
- 3historical official price for this version; current rolling-model pricing may differ
- 4standard tier; priority and flex tiers differ
- 5cheapest of its six regional price tables
- 6cheapest input-length tier; a longer prompt costs more per token
- 7limited-time promotional rate; the list price is higher
- The cash column is the credit rate divided by 10, the top-up rate.
- "vs vendor official" compares against the model vendor's own published list price, taking their cheapest published tier — DeepSeek's off-peak rate, Alibaba's cheapest region and shortest input tier, and cache-miss pricing throughout. Aggregator prices are not used here; they are on the comparison page.
- OpenAI uses Standard processing with up to 272K input tokens; GPT-5.6 Sol uses its current promotional rate.
- Not every model has a comparable figure. 2 of the 31 are not on their vendor's own rate card. A blank cell means we found no source we could cite, not that the model is cheap.
- Free models are listed too, and carry no multiple: their cash price is $0, so there is no ratio to compute. The figure in that column is the vendor's own list price.
- Free status is set by our upstream provider and can be withdrawn — when it is, these calls start billing at the listed rate and your code does not change. The models page always shows the current state.
- Vendor APIs bill cached input far below their cache-miss rate, and we have no equivalent tier. On workloads that hit the cache constantly, the vendor can be cheaper than us — the comparison page has the numbers.
Up front
"$1 becomes $10" is credit, not a discount coupon
You pay $1 and your account gains $10 in credit. Credit burns at each model's multiplier, which varies widely — so it is not "official pricing at 10% off". Treat the actual deductions in the request log as the real cost.
Invitations
Bring someone with you and you both get credit
Every account has an invitation link in the console. When someone signs up through it, you and they each receive $1 in credit, once. Sign-up credit still applies on top, so an invited account starts with $2 in credit — about 20M input tokens on openai/gpt-6-luna — without adding a card.
Start now
Sign in and your account opens automatically with a default key and $1 in credit. No card, no forms.