Pricing
Buy a credit pack once, spend it as you generate images. Failed jobs are refunded automatically.
Credits per image by model
One key, one balance, every model. Cost scales with resolution and the number of images requested (n, up to a model's limit). Failed jobs are refunded in full.
| Model | 1K long edge ~1024 px | 2K long edge ~2048 px | 4K long edge ~4096 px |
|---|---|---|---|
nanobanana Google's fast Gemini image model: general-purpose text-to-image and image-to-image at 1K with reference-image input — the lowest price per image in the catalogue. up to 4 images per request | 2credits ≈ $0.025 / image | — | — |
nanobanana-pro Google's higher-fidelity Gemini image model: more detail and closer prompt adherence, with 1K, 2K and 4K output for hero shots and print-sized crops. up to 4 images per request | 4credits ≈ $0.05 / image | 6credits ≈ $0.075 / image | 8credits ≈ $0.10 / image |
qwen-image-2 Alibaba's Qwen-Image 2.0: strong at rendering legible text inside images, including Chinese, and at illustration styles; 1K and 2K output with reference-image input. up to 4 images per request | 3credits ≈ $0.038 / image | 4credits ≈ $0.05 / image | — |
qwen-image-max The flagship of Alibaba's Qwen-Image line: the highest-fidelity tier for photoreal and text-heavy compositions, with 1K and 2K output and reference-image input. up to 4 images per request | 8credits ≈ $0.10 / image | 9credits ≈ $0.113 / image | — |
seedream-5 ByteDance's Seedream 5.0: a high-resolution model for photoreal frames and in-image text, served at 2K and 4K, with reference-image input. up to 4 images per request | — | 16credits ≈ $0.20 / image | 18credits ≈ $0.225 / image |
wan-2-7-image Alibaba's Wan 2.7 Image (Tongyi Wanxiang): fast, low-cost text-to-image and reference-image generation at 1K and 2K — for previews and bulk generation. up to 4 images per request | 2credits ≈ $0.025 / image | 3credits ≈ $0.038 / image | — |
Pick a tier with resolution and a shape with ratio. Money figures use the best-value pack ($0.013 per credit). Cost = credits × n; failed jobs are refunded in full.
Credits per 1M tokens by model
Chat models are billed per token — input (the messages you send) and output (the reply, including reasoning) at separate rates. Credits are reserved when the request starts and settled to the actual usage when it ends; a decimal credit is charged exactly, so a short request costs a fraction of a credit.
| Model | Context | Tools | Input / 1M | Output / 1M |
|---|---|---|---|---|
doubao-seed-2-pro ByteDance's flagship Doubao model with vision and web search — for complex reasoning and multimodal tasks. | 256K | yes | 85credits $1.063 ≈ $0.0011 / 1K | 408credits $5.10 ≈ $0.0051 / 1K |
doubao-seed-2-code Doubao's coding-tuned model with vision — for agentic coding, refactors and UI-from-screenshot tasks. | 256K | yes | 85credits $1.063 ≈ $0.0011 / 1K | 408credits $5.10 ≈ $0.0051 / 1K |
doubao-seed-2-lite Mid-tier Doubao model with vision — general assistants and image-aware chat at a lower cost than Pro. | 256K | — | 17credits $0.213 ≈ $0.0002 / 1K | 85credits $1.063 ≈ $0.0011 / 1K |
doubao-seed-2-mini The cheapest Doubao model with vision — high-volume, latency-sensitive chat and simple image tasks. | 256K | — | 2.146credits $0.027 ≈ $0.00 / 1K | 68credits $0.85 ≈ $0.0009 / 1K |
kimi-k3 Moonshot's flagship Kimi model with a huge context window — for very long documents and multi-step agents. | 1M | yes | 197.028credits $2.463 ≈ $0.0025 / 1K | 985.815credits $12.323 ≈ $0.0123 / 1K |
claude-opus-5-5 Anthropic's top Claude Opus model — strong reasoning and agentic coding with a large context window for complex, multi-step work. | 200K | yes | 146.88credits $1.836 ≈ $0.0018 / 1K | 734.4credits $9.18 ≈ $0.0092 / 1K |
claude-fable-5-1 Anthropic's premium Claude model for long-form writing — polished drafting and editing with a large context window. | 200K | yes | 1,700credits $21.25 ≈ $0.0213 / 1K | 1,700credits $21.25 ≈ $0.0213 / 1K |
gpt-6-astra OpenAI's general-purpose flagship — balanced reasoning, chat and tool use for everyday production assistants. | 128K | yes | 68credits $0.85 ≈ $0.0009 / 1K | 340credits $4.25 ≈ $0.0043 / 1K |
gpt-5.6-terra OpenAI's fast, economical GPT model for high-volume chat, drafting and tool use where cost and latency matter most. | 128K | yes | 17credits $0.213 ≈ $0.0002 / 1K | 170credits $2.125 ≈ $0.0021 / 1K |
qwen3.8-max Alibaba's newest flagship Qwen model, with a very large context window and solid tool use — a sensible default for most agent and reasoning work. | 1M | yes | 138.718credits $1.734 ≈ $0.0017 / 1K | 416.161credits $5.202 ≈ $0.0052 / 1K |
Credits per 1M tokens; every message in the request is input, the reply (including reasoning) is output. Money figures use the best-value pack ($0.013 per credit). Nothing is charged when no tokens were produced.
How to pay
Packs are prepaid: the balance is shared by every model and does not expire. Payment methods available right now:
- Bank card or СБП (Faster Payments) in roubles — the receipt is sent to your email.
- Cryptocurrency — USDT and other supported coins — through a hosted invoice; credits are added once the network confirms the transfer.
- Can I pay with a Russian bank card or СБП?
- Yes. Choose a pack, pick card or СБП at checkout and pay in roubles at the price shown on this page. Credits are added as soon as the payment is confirmed, usually within seconds, and the receipt goes to your email.
- Can I pay with crypto (USDT)?
- Yes. Choose a pack and pay the invoice in USDT or another supported coin. Credits are added automatically after the network confirms the transfer, usually within a few minutes.
- Are unused credits refundable?
- Unused credits are refunded on request to the original payment method within 10 working days. Credits spent on successful generations are not refundable; failed jobs are refunded to the balance automatically.
Billing
How credits are spent on images and on tokens.
- How is a chat request billed?
- Per token, at the model's input and output rates in credits per 1M tokens. Every message you send — system, user and earlier assistant turns — counts as input; the reply counts as output. Worked example: on a model priced at 20 credits per 1M input and 40 per 1M output, a request with 2,000 prompt tokens and a 500-token reply costs 2,000 ÷ 1,000,000 × 20 + 500 ÷ 1,000,000 × 40 = 0.04 + 0.02 = 0.06 credits. The usage block of every response shows the token counts and the cost.
- Can a request cost a fraction of a credit?
- Yes. Token-billed requests are charged in decimal credits with up to three decimal places, so a short chat request costs a few thousandths of a credit rather than being rounded up to one. Image generations stay whole credits per image. Your balance and the ledger show the exact figures.
- What am I charged when a chat request fails or I abort a stream?
- Nothing when no tokens were produced: a rejected request, a temporarily unavailable model or an upstream failure refunds the reservation in full. Credits are reserved for the estimated prompt plus the reply cap and settled to the actual usage once the call ends. If you close a stream early you pay only for the tokens generated up to that point.
- Do thinking (reasoning) tokens cost extra?
- They are billed as output tokens at the model's output rate — nothing more. When you enable thinking on a model that supports it, the reasoning text arrives alongside the answer and its tokens appear in completion_tokens; they also count toward max_tokens, so set it high enough for the task.