Starter credit Web
$5 one-time source (opens in a new tab)
One-time
For: Trying the endpoint before subscribing
Limits
- No subscription
- No recurring charges
Includes
- $5 usage credit
API included
CodingDocumented overviewLast checked Sep 29, 2026
Private inference endpoint for open-weight models, built for coding agents, with zero request retention.
(open full size in a new tab)
(open full size in a new tab)
(open full size in a new tab)
(open full size in a new tab)
(open full size in a new tab)
(open full size in a new tab)Our take: Choose Zro if you run coding agents such as Claude Code or Codex on open-weight models and need a stated zero-retention, no-training endpoint with EU or US routing.
Main limitation: Monthly plans cover estimated token totals, not fixed allowances, and agent tool calls (web search, fetch) are billed per call on top.
Choose Zro if you run coding agents such as Claude Code or Codex on open-weight models and need a stated zero-retention, no-training endpoint with EU or US routing. Based on the vendor's documentation, not on a hands-on test.
Also a good fit for
Or if
Prices checked Sep 29, 2026 on the official US pricing page. Official pricing ↗ Prices and limits change often.

No free tier; a one-time $5 usage credit needs no subscription.
$5 one-time credit.
$5 one-time source (opens in a new tab)
One-time
For: Trying the endpoint before subscribing
Limits
Includes
API included
$20/mo source (opens in a new tab)
Month, annual commitment
$16/month billed yearly
For: Everyday coding-agent work
Limits
Includes
API included
$60/mo source (opens in a new tab)
Month, annual commitment
$48/month billed yearly
For: Heavier, longer-running agent work
Limits
Includes
API included
$5 one-time source (opens in a new tab)
One-time
For: Top-ups beyond the monthly budget
Limits
Includes
API included
Custom pricing source (opens in a new tab)
Unknown
For: Teams needing processor terms
Limits
Includes
Agent tool calls are billed per call: web search $0.02/$0.20, fetch $0.01/$0.02.
Not stated.
Zro is an inference service from MoonMath that serves open models such as GLM, Kimi and DeepSeek through an OpenAI-compatible API and a Messages API. A CLI launches coding agents like Claude Code and Codex against it. The site states zero request retention and no training on customer requests.

You provide
You get back
Scenarios are worked out by our editors from the documentation; the actual result depends on your input and plan.

Accepts chat-completions/Responses calls at /v1 and Anthropic-style Messages requests at /v1/messages. Lets existing clients and agent tools switch by changing the base URL.
npm package @moonmath-ai/zro logs in once and launches Claude Code, Codex, OpenCode, Hermes, OpenClaw or Pi with temporary session config. Avoids hand-editing each agent's config.
Each API key sets inference to Europe, the United States or either region. Keeps request content in an allowed region.
Caches repeated prompt prefixes so long agent sessions cost fewer tokens. Cuts cost and latency in multi-turn work.
Per-call web search with source URLs and URL fetch returning markdown. Gives agents live web context.

Refactor the auth module to remove duplicated token checks and run the tests.

Zro (by MoonMath) has no public API in its official documentation. Base URL https://zro.moonmath.ai/v1; Messages at /v1/messages API documentation ↗
Summarized from the vendor's current policies; this is not legal advice and a vendor's badge is not an independent audit.
Prompts and completions are processed transiently to answer the request.
Not used for training, fine-tuning or evaluation without explicit agreement.
Prompts, responses and request bodies not retained; content logging disabled at gateway.
Not stated on reviewed pages.
DPA mainly for Team and Enterprise.
Policies: zro.moonmath.ai/privacy
"Better suited" is based on documented features and pricing, not on a head-to-head test.
| Alternative | When to choose it | Pricing | Limitation | |
|---|---|---|---|---|
| Claude | — | — | ||
| Cursor | — | — | Compare | |
| GitHub Copilot | — | — | Compare | |
| Devin Desktop (formerly Windsurf) | — | — | Compare |
No user reviews yet.
Used Zro (by MoonMath)? Be the first to share your experience.
GLM-5.3 and GLM-5.3 Flash, DeepSeek V4.1 Flash, Kimi K3 and Dolly 1 Security are listed; availability is shown per region when you create a key. source (opens in a new tab)
No. The privacy notice says prompts, responses and request bodies are not retained and there is no setting that enables retention. source (opens in a new tab)
Yes. Each API key allows Europe, the United States or either; all requests still pass through the EU gateway for authentication. source (opens in a new tab)
No. The ~1B and ~5B figures are expected totals based on typical cached agent usage, not fixed allowances. source (opens in a new tab)
Claude Code, Codex CLI, OpenCode, Hermes, OpenClaw and Pi; Cursor and Cline use manual setup per the integrations docs. source (opens in a new tab)
Not verified: Supported countries not seen; Team and Enterprise prices; Commercial-use terms.
Every fact about Zro (by MoonMath) above comes from the 4 official pages listed here, last compared on Sep 29, 2026; we have not run Zro (by MoonMath) ourselves, so it has no editorial score. How we check and re-check pages.
Article updated Oct 10, 2026; prices checked Sep 29, 2026. Outbound links to vendors are plain links; we do not earn commissions. How we review · Report an error
compare>