Pricing

Claude at 30%, GPT at 1/30 of official, billed per-token

No monthly fee, no plan, no minimum top-up. Top up what you want, use what you need. When the upstream raises prices we follow up; when they cut we follow down. The ratio doesn't move.

Claude Opus 4.7 · Opus 4.6 · Sonnet 4.6 · Haiku 4.5 all at × 0.30 (3折, 70% off). GPT supported (codex / gpt-5.5, as low as 1/30 of official); Gemini coming soon.
How to pay

Same price across methods. Pick what's easy for you.

Price is fixed — payment method doesn't change it. The cards below just describe the onboarding flow and who each fits.

WeChat / Alipay

¥ scan to pay

Solo developers, side projects, small tools — fastest path to start.

  • Telegram payment slip; scan to pay = instant
  • Any amount, $10 minimum
  • FX at the day's USD mid-rate
  • Refundable balance
Top up via Telegram
Recommended

USDT (TRC20 / ERC20)

USDT on-chain

Technical teams, overseas projects, FX-sensitive customers.

  • 1:1 USD settlement, no FX spread
  • TRC20 / ERC20; on-chain confirmation = credit
  • $500+ top-ups get priority support
  • No Chinese bank account required
Get USDT address

Wire transfer

¥ $1000+/mo

Companies needing fapiao, monthly settlement, dedicated rep.

  • Company-to-company wire, T+1 settlement
  • 普票 / 专票 issued on demand
  • Monthly settlement; annual frame deals negotiable
  • Dedicated account manager + WeChat group
Discuss wire
Model Pricing

Per-model pricing, by provider

All prices per 1M tokens, in USD. Claude price = Anthropic official × 0.30 (3折); GPT as low as 1/30 of official. The strikethrough shows the official list price so you can reconcile.

Anthropic · Claude (official × 0.30, 3折 · 70% off)
Official list from platform.claude.com/docs/en/about-claude/pricing. Cache Write column shows the 5-minute cache rate (1.25× official input). Cache Read is 0.1× official input. The Batch column includes Anthropic's 50% batch discount stacked with our × 0.30.
Model Input $/MTok Output $/MTok Cache Write (5m) Cache Read Batch (In / Out)
claude-opus-4-7 $15.00$4.50 -70% $75.00$22.50 -70% $18.75$5.63 $1.50$0.45 $7.50/$37.50$2.25 / $11.25
claude-opus-4-6 $15.00$4.50 -70% $75.00$22.50 -70% $18.75$5.63 $1.50$0.45 $7.50/$37.50$2.25 / $11.25
claude-sonnet-4-6 $3.00$0.90 -70% $15.00$4.50 -70% $3.75$1.13 $0.30$0.09 $1.50/$7.50$0.45 / $2.25
claude-haiku-4-5 $1.00$0.30 -70% $5.00$1.50 -70% $1.25$0.38 $0.10$0.03 $0.50/$2.50$0.15 / $0.75
OpenAI · GPT / codex LIVE
OpenAI upstream is wired, as low as 1/30 of official (~3.3%). codex / gpt-5.5 lineup, billed per-token with the same budget / rate-limit system as Claude. codex CLI uses /v1/responses (base_url=…/v1, wire_api=responses). Full GPT model list + per-model unit prices are kept in sync with OpenAI's official table and announced in the Telegram group.
Google · Gemini COMING SOON
Gemini upstream is also on the roadmap. We don't quote placeholder prices for things not yet wired — Gemini rows will appear in this table only once the upstream is real.

Claude prices track Anthropic's official list; discount ratio stays at 0.30 (3折); GPT as low as 1/30 of official. Last sync: 2026-05-02 (Claude Opus 4.7 in table).

Worked Examples

Three real-world monthly bills

Model × monthly volume = bill. The "SACTL monthly" line is what you actually pay — no hidden markups.

A · Customer support (Sonnet 4.6)

144,000 req/mo · 500 in + 300 out each · Q&A, form parsing, multi-turn context

Input · 72M × $0.90/M$64.80
Output · 43.2M × $4.50/M$194.40

SACTL monthly

$259.20

Official list: $864.00

Save $604.80/mo (70%)

B · Opus review (Opus 4.6)

3,000 req/mo · 20K in + 2K out · Long-doc compliance review, financial risk

Input · 60M × $4.50/M$270.00
Output · 6M × $22.50/M$135.00

SACTL monthly

$405.00

Official list: $1,350.00

Save $945.00/mo (70%)

C · High-frequency light tasks (Haiku 4.5)

1,000,000 req/mo · 500 in + 200 out · Bulk classification, tagging, realtime completions

Input · 500M × $0.30/M$150.00
Output · 200M × $1.50/M$300.00

SACTL monthly

$450.00

Official list: $1,500.00

Save $1,050.00/mo (70%)

FAQ

The questions we get most

1. Really 30% for Claude and 1/30 for GPT? Any monthly fee or starter pack?

Claude is 30% across the board (× 0.30) — Opus / Sonnet / Haiku, same ratio, no usage tiers; GPT (codex / gpt-5.5) is as low as 1/30 of official. No monthly fee, no plan, no minimum top-up — $10 works. Billed per token. Each response header reports x-sactl-usage-cost-usd so you can reconcile with a calculator.

2. How do I top up, and how fast does it land?

WeChat / Alipay: Telegram support sends a payment slip; scan to pay = instant credit. USDT: we send a wallet address; on-chain confirmation (~1 min on TRC20, a few minutes on ERC20) = credit. Wire transfer: we send the bank info; T+1 settlement; annual frame deals possible. FX is settled at the day's USD mid-rate — no spread on top.

3. Can you issue Chinese fapiao (普票 / 专票)?

Yes, both 普票 (general) and 专票 (special VAT). WeChat / Alipay top-ups above $50 are eligible for 普票; wire transfers can get 专票 with company name, tax ID, and bank info. Invoice content is usually "技术服务费 / 软件服务费" — adjustable for your finance department. Send the request to Telegram support.

4. How do Claude Code / Codex CLI / Cursor / SDK integrate?

Claude Code: set ANTHROPIC_BASE_URL=https://api.sactl.ai + ANTHROPIC_AUTH_TOKEN=sk-xa-*. Codex CLI: add a custom provider in ~/.codex/config.toml with base_url=https://api.sactl.ai/v1, wire_api=responses, env_key=SACTL_API_KEY. OpenAI SDK: set base_url to https://api.sactl.ai/v1, swap the API key to sk-xa-*. Cursor: in settings, set Custom OpenAI Endpoint to the SACTL URL. Full guide: docs.

5. Is access from China stable? Will it suddenly drop?

Singapore + Hong Kong dual-route, ~80ms average latency. Multi-account upstream pool: when one upstream account hits risk control, our backend swaps to the next; the client sees a 200. We can't dodge global Anthropic outages — but we post status to the Telegram group when they happen. We can share runtime history on request.

6. Can I get a refund? What if you shut down?

Balance is refundable on request at any time (after deducting what you've used) — message Telegram support. If we ever shut down, we give 30 days notice and refund all balance pro-rata to the original payment channel. We do not sell prompts, do not store request bodies, and do not resell your data. That's the floor.

Ready to use Claude at 30% and GPT at 1/30?

Message us on Telegram and get your first sk-xa-* in minutes — works with Claude Code and Codex CLI out of the box.