What we do
ThunderPeak is an independently operated AI API gateway. We aggregate access to leading model families — Anthropic's Claude, OpenAI's GPT, DeepSeek, Alibaba's Qwen, Moonshot's Kimi and Zhipu's GLM — behind a single API key, a single balance and two industry-standard wire formats: OpenAI-compatible chat completions and Anthropic-compatible messages.
In practice that means you change one line of configuration — the base URL — and your existing tools keep working. Claude Code, Codex CLI, Cursor, Cherry Studio, VS Code extensions and any official OpenAI or Anthropic SDK can all point at the same gateway and share the same prepaid credits.
Why we exist
Working with multiple model providers today is needlessly painful:
- Accounts multiply. Every provider wants its own signup, its own verification, its own dashboard.
- Payments get complicated. Overseas cards fail, minimum top-ups are high, and currencies differ.
- Bills fragment. Usage is scattered across portals that never talk to each other.
- Switching costs are real. Every new model means new SDKs, new endpoints, new failure modes.
ThunderPeak collapses all of that into one key, one balance and one compatible API surface — so your team spends its time building, not maintaining provider plumbing.
How we price
We sell access to Claude and GPT at 66% of official rates, DeepSeek and Kimi K3 at 88%, GLM-5.3 at 80% and GLM-5.2 at 70%, and Qwen at official rates. These are standing prices, not launch promotions — the discount you see today is the discount we intend to keep.
Credits are prepaid and never expire. There is no subscription, no seat count, no minimum monthly commitment. You top up with USDT on Tron (TRC-20), BSC or Solana from $10, and every request is metered at the published per-model token rates with input, output, cache-write, cache-read and thinking tokens accounted for individually.
Everything is visible: model prices, token-level usage, per-key limits and request history live in your console. If a number surprises you, you can trace exactly which request produced it.
How it works under the hood
Requests arrive at our edge, are normalised across protocols, routed to the fastest healthy upstream for the requested model, and streamed back to you as standard SSE. Keys are isolated per customer with independent rate limits and spending controls, and upstream failures retry automatically before surfacing an error to your application.
That architecture is also why one key can serve a hobby project, a production agent fleet and a reseller's customer base without interference — each key gets its own quota, limits and usage record.
Who uses ThunderPeak
Independence & compliance
ThunderPeak is an independent service. We are not affiliated with, endorsed by, or acting as an official channel for Anthropic, OpenAI, DeepSeek, Alibaba, Moonshot or Zhipu. Model names are used solely to describe API compatibility. If you build a product on top of us, please state the same to your own users.
We ask the same of everyone using the gateway: no abuse, no fraud, no misrepresentation. And we reserve the right to suspend accounts or keys that harm the service or other customers.
Support
Support is human and responsive. Reach us on Telegram @thunderpeak_support or by email at [email protected] — we reply in your language. There is also an AI assistant on every page of this site for quick questions about pricing, models and top-ups.
The name
"ThunderPeak" is about speed meeting summit: the sharpest models, one strike away. We kept the product deliberately plain — clear prices, clear limits, no dark patterns — because infrastructure should be boring in the best way.