Compute credit packs
Start with ¥99. You are not asked to prepay a large amount up front — first verify Base URL, API key, model calls, and the credit ledger. After that works, move to ¥499 / ¥999 or a project plan. Text bills by usage; images bill per successful generation. Failed requests are usually not charged — Usage and Credits are authoritative.
This page covers pricing only. Capabilities live on Models; curl examples live on Docs. Fits Cherry Studio / Cursor / OpenAI SDK / API projects.
Start with ¥99 to run the path end-to-end. Reference prices help planning — actual charges follow Usage and the compute-credits ledger.
Recommended packs
¥99 to prove the integration; ¥499 for small teams; ¥999 for higher prepaid balance. Self-serve packs are for onboarding and mid-scale use — not a formal large-buyer SLA.
Recommended — API trial pack
Compute credits work for Chat API and Image API.
Teams & steady daily use
Compute credits work for Chat API and Image API.
Higher prepaid balance — contact us for heavy production
Compute credits work for Chat API and Image API.
Smaller top-ups
¥10 / ¥20 / ¥49 remain for light smoke tests. Most projects should start at ¥99.
Best for first top-up / smoke tests
Compute credits work for Chat API and Image API.
Best for light usage
Compute credits work for Chat API and Image API.
Best for regular usage
Compute credits work for Chat API and Image API.
WeChat Pay and Visa / international cards are supported.
After purchase, check the compute-credits ledger for balance and history. Test in Chat or Image Playground.
Usage-based billing
Prepaid credits power every API call. You only pay for what you use.
- Recommended: ¥99 / ¥499 / ¥999. Smaller ¥10 / ¥20 / ¥49 packs remain for light tests (¥1 = 10,000 credits).
- Compute credits work for Chat API and Image API.
- Successful calls debit credits.
- Failed calls are usually not charged — ledger is authoritative.
- Chat models: billed by input/output tokens.
- Image models: billed per generation in credits.
- Monitor usage in Usage.
- Ledger in Credits.
Model pricing
10,000 compute credits ≈ ¥1 base. Chat API bills by input/output tokens. Image API bills per generation. Usage and Credits are the source of truth for actual charges.
Current reference prices are shown for planning. Usage and Credits are the source of truth for actual charges. Failed requests are usually not charged.
Chat Models
| Model | Model ID | Input | Output | Billing unit | Tags |
|---|---|---|---|---|---|
| GPT 5.4 | gpt-5.4 | ¥0.7~¥1.4 / 1M tokens | ¥6~¥12 / 1M tokens | Billed by input/output tokens | — |
| GPT 5.5 | gpt-5.5 | ¥2.2~¥4.4 / 1M tokens | ¥13.5~¥27 / 1M tokens | Billed by input/output tokens | — |
| Gemini 3 Flash | gemini-3-flash | ¥0.4~¥0.8 / 1M tokens | ¥3~¥6 / 1M tokens | Billed by input/output tokens | Flash |
| Gemini 2.5 Flash | gemini-2.5-flash | ¥0.3~¥0.6 / 1M tokens | ¥2~¥4 / 1M tokens | Billed by input/output tokens | Flash |
| Gemini 3 Pro | gemini-3-pro | ¥1.5~¥3 / 1M tokens | ¥7~¥14 / 1M tokens | Billed by input/output tokens | — |
| Gemini 2.5 Pro | gemini-2.5-pro | ¥1.25~¥2.5 / 1M tokens | ¥6.25~¥12.5 / 1M tokens | Billed by input/output tokens | — |
Image Models
Image generation is billed per successful generation in credits, with a current reference price in ¥. Each model uses a different amount; failed requests are usually not charged.
| Model | Model ID | Credits price | Reference price | Billing unit | Tags | Best for |
|---|---|---|---|---|---|---|
GPT Image 2 | gpt-image-2 | 600 compute credits / generation | ~¥0.06 / generation | Per generation | — | Compatible-style image model (async Image API only) |
GPT Image 2 VIP | gpt-image-2-vip | 1,300 compute credits / generation | ~¥0.13 / generation | Per generation | VIP | VIP compatible-style image model (async Image API only) |
Nano Banana Fast | nano-banana-fast | 440 compute credits / generation | ~¥0.04 / generation | Per generation | Fast | Lightweight fast images / lower cost (async task_id) |
Nano Banana | nano-banana | 1,400 compute credits / generation | ~¥0.14 / generation | Per generation | — | Recommended image model (async task_id; bill on success) |
Nano Banana 2 | nano-banana-2 | 1,200 compute credits / generation | ~¥0.12 / generation | Per generation | — | Higher quality / more stable (async task_id) |
What scale is Tokfai for?
Self-serve packs fit developers and small teams: Cherry Studio / Cursor, small tools and scripts, ecommerce titles / details / image prompts, engineering materials and log tidy-up, and a ¥99 first pass on API speed and cost.
¥99 / ¥499 / ¥999 are for proving the path and running small-to-mid volume. They are not a high-concurrency production contract.
If you need frequent calls, batch jobs, production wiring, or engineering project work, email junpengpanchina@gmail.com first to review RPM, QPS, tokens, models, use case, and budget.
Tokfai does not currently take million-RPM workloads, model distillation, training-data supply, large-scale scraping, grey-market / illegal use, formal large-buyer SLA procurement, or personal privacy / sensitive data processing as self-serve offers.
Review items include expected RPM, peak concurrency, daily requests, average tokens, model types, whether image jobs are involved, whether it is production, and whether contract / invoice / written commercial terms are required.
Contact project planPricing FAQ
- What is the ¥99 pack?
- ¥99 API trial pack credits about 1,188,000 compute credits (+20% bonus). Same balance for Chat and Image — use it to prove Base URL, key, models, and the ledger before larger top-ups.
- Who is Tokfai for?
- Developers and small teams using Cherry Studio, Cursor, OpenAI SDK, or custom apps; ecommerce copy/image prompts; engineering notes and logs. Start with ¥99 to verify API, speed, and cost.
- Who should not self-serve first?
- Million-RPM, distillation, training-data supply, large-scale scraping, grey-market / illegal use, formal SLA procurement, or personal privacy / sensitive data workloads. Early public beta — email junpengpanchina@gmail.com if you need a human review.
- Can I buy ¥999 and run production at high concurrency?
- No automatic promise. Self-serve packs only credit your balance; high-frequency or production use needs a capacity review. Limits and Usage / Credits ledger are authoritative.
Related pages
Pricing stays separate from integration tutorials. Jump to Models or Docs as needed.
Base URL: https://api.tokfai.com/v1 · Key: sk-tokfai_... · Successful calls debit compute credits. Failed calls and image timeouts are usually not charged; Usage and Credits ledger are authoritative.