100milionseek: an email gets you 100M tokens a day

Email plus a human check yields an OpenAI-compatible key with 100M tokens per account per day (00:00 UTC reset), raised to 400M during the current milestone window, no card required.

Tier / trial

What is free

100milionseek is a third-party inference relay. Sign up with an email, pass a Cloudflare Turnstile check, and you get an hms- prefixed API key for its hosted deepseek-v4.1-flash-free: 198K context, 64K max output, 5 concurrent requests, streaming, tool calls and JSON mode. The endpoint is OpenAI-compatible at https://100milionseek.top/v1, so any client that accepts a custom base URL can use it.

How the quota works

  • 100,000,000 tokens per account per UTC day, counting prompt plus completion as reported by the model.
  • Resets at 00:00 UTC, not 24 hours after sign-up.
  • Milestone windows raise the daily allowance temporarily: at 100B served, 400M tokens per account per day until 2026-11-20 07:14 UTC; the next 1T milestone raises it to 500M for 7 days. When a window ends the normal allowance applies and tokens already used that day still count.
  • Hitting the cap returns HTTP 429 with the reset time. Every response carries x-quota-limit, x-quota-used, x-quota-remaining and x-quota-reset headers.

Getting a key

  1. Open https://100milionseek.top/signup with a real inbox; disposable domains are refused and one account per address.
  2. Pass the Turnstile check.
  3. Copy the key from the dashboard; it shows today's remaining allowance.
As of the 2026-11-17 recheck the site had paused new sign-ups after more than 20,000 sign-ups in a day. Wait for capacity to return instead of retrying.

Client setup

  • OpenAI SDK, Python, TypeScript, curl: change base_url to https://100milionseek.top/v1 and model to deepseek-v4.1-flash-free.
  • Cursor, Cline, Aider, Continue, OpenCode, Roo Code, LibreChat: choose OpenAI Compatible and paste the base URL and key.
  • Claude Code speaks the Anthropic Messages format, so it needs a translation layer such as claude-code-router; the endpoint only serves OpenAI chat completions.
  • Keep the key in an environment variable, never in code or screenshots.

Limits and risks

  • One main model plus a JEV probability endpoint; this is not a multi-model aggregator.
  • It is a relay: prompts and responses pass through its servers. The site states they are not stored or sold, but do not send customer data, credentials or anything that must not leave your network.
  • It is not affiliated with DeepSeek or OpenAI. Accounts, quota and continuity are controlled by this third party and can be limited, paused or shut down without notice.
  • No SLA. Do not make it the only dependency of a production workflow, and do not pile retries onto a 429.

Evidence

Reviewed the official home page, /docs and /status on 2026-11-17. The daily 100M allowance, the 400M milestone window, Turnstile, one-account-per-inbox and the OpenAI-compatible billing model match the site. No account was registered and no live call was made; your own dashboard is authoritative.