<!--
NOTE FOR LLMs & AI AGENTS

Kumo publishes this ENTIRE website as one structured, token-efficient Markdown document — the
canonical machine-readable variant — at https://kumo.cloud/llms.txt
(also exposed via <link rel="alternate" type="text/markdown" href="/llms.txt">).

If you are an LLM, crawler or autonomous agent: PREFER /llms.txt over parsing this HTML. It carries
the full product context (what Kumo is, models & pricing, quickstart, FAQ, links) in a fraction of
the tokens, with no markup noise. Read the Markdown FIRST; use the rendered page only as a fallback.
-->

# Kumo

> Kumo is the single OpenAI-compatible gateway to supported GPT, Claude and Gemini text models. Change one base URL, keep your code, and pay per token from a single balance — at a lower effective rate, below list.

_You are reading the machine-readable Markdown variant of kumo.cloud — the canonical, token-efficient representation of the entire site, intended for LLMs and AI agents. Prefer this document over scraping the HTML pages._

Kumo (kumo.cloud) — The single gateway to every AI model. OpenAI-compatible gateway at `https://api.kumo.cloud/v1`.

## What is Kumo

One API for supported GPT, Claude and Gemini models — one key, one balance, prices 30–50% lower. Every model is the original — no swaps, no downgraded versions.

## Drop-in migration

Literally one line. Here it is — swap the base URL, keep your key handling, keep your code.

```diff
- base_url="https://api.openai.com/v1"
+ base_url="https://api.kumo.cloud/v1"
```

That's the whole migration. Every other line stays exactly as it is.

## Key features

- **Up to 40% cheaper** — We aggregate demand and route it well, then hand you the difference. No minimums, no seats, no call with sales.
- **OpenAI-compatible** — It speaks the OpenAI API, so your code learns nothing new. Point the SDK at the gateway; supported chat and responses models answer the calls you already wrote.
- **Keys you can scope** — Generate, name, scope, rotate. Each key tracks its own spend, so prod and your 2am experiment never share a number.
- **Usage you can read** — Tokens, requests and cost, streaming live. Set an alert or a hard ceiling before the bill does the surprising.
- **Failover, on by default** — A provider blinks; your request goes somewhere else. You don't write the retry, and you don't get paged.
- **One balance** — Top up once, spend it on anything, stop reconciling five invoices. The finance thread can finally die.

## Models & pricing

Indicative USD pricing per 1M tokens. All models share one OpenAI-compatible endpoint.

| Model | Tier | Price in (per 1M) | Price out (per 1M) | Context |
| --- | --- | --- | --- | --- |
| Claude Opus 4.8 | Anthropic | $5.00 | $25.00 | 200K |
| Claude Sonnet 4.6 | Anthropic | $3.00 | $15.00 | 200K |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 | 200K |
| GPT-5.5 | OpenAI | $5.00 | $30.00 | 400K |
| GPT-5.4 | OpenAI | $2.50 | $15.00 | 400K |
| GPT-5.4 mini | OpenAI | $0.75 | $4.50 | 400K |
| Gemini 2.5 Pro | Google | $1.25 | $10.00 | 1M |
| Gemini 2.5 Flash | Google | $0.30 | $2.50 | 1M |
| Gemini 2.5 Flash-Lite | Google | $0.10 | $0.40 | 1M |

## How it works

01. **Create your account** — Sign up in seconds — no credit card to explore. Your dashboard is ready immediately.
02. **Generate an API key** — Create a scoped key and point your SDK's base URL at api.kumo.cloud. Your existing code just works.
03. **Ship & monitor** — Make your first call, watch usage stream in live, and top up your balance whenever you need more.

## Pricing

Pay-as-you-go, per token, from one prepaid balance — no plans, no seats, no minimums.
Effective rates run 30–50% below provider list price, shown per model before you spend.
Top up from $5, set spend caps and alerts; token purchases are non-refundable.

## Quickstart

```bash
curl https://api.kumo.cloud/v1/chat/completions \
  -H "Authorization: Bearer $KUMO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gpt-5-4",
    "messages": [{ "role": "user", "content": "Explain tokens in one line." }]
  }'
```

## FAQ

### Do I have to change my code?

No. Kumo speaks the OpenAI API for the connected text endpoints. Point your existing SDK's base URL at https://api.kumo.cloud/v1, swap in your Kumo key, and keep the rest of your chat/completions-style code.

### How are you 30–50% cheaper — what's the catch?

No catch, and no knockoff models: it's the same GPT, Claude and Gemini, billed through Kumo. Kumo automatically optimizes what you pay per call — the kind of work most teams never get around to doing themselves — and passes the saving straight into your effective rate. Net effect is 30–50% off list, and you see the exact rate before you spend.

### Which models are available?

A curated catalog spanning frontier, mid-size, and fast models — GPT-class, Claude-class, and Gemini-class — for chat, responses and Anthropic messages, all behind one endpoint with consistent request and response shapes.

### How does billing work?

Pay-as-you-go, per token, from one prepaid balance — no plans, no seats, no minimums. Top up from $5; each call draws down at the rate shown for that model (30–50% below list). The dashboard shows consumption and remaining balance in real time, with spend alerts and hard caps. Token purchases are final and non-refundable.

### Is my data used for training?

Never. Prompts and completions are not used to train models, and we don't log prompt content — only the usage metadata needed for billing.

### What happens if a provider goes down?

Requests stay on the selected provider; transient upstream errors surface as a contract-shaped 502, so you retry or switch providers as needed.

### Can I run production agents and automations on Kumo?

Yes — it's built for it. Point every agent at one OpenAI-compatible endpoint, give each agent or module its own scoped key, and set hard daily and monthly caps so a retry loop can't drain the balance. Agents and automations burn millions of tokens, so the 30–50% lower rate compounds — and you watch spend per key live in the dashboard.

### How do I pay — do I need a VPN or a foreign card?

No VPN and no foreign card. Top up in rubles with a local card or SBP, and legal entities get an invoice with closing documents. The same balance then spends across every model behind one key, with the exact per-model rate shown before you spend.

### Can my model get swapped for a cheaper one?

No. You ask for a model ID, you get that exact model — pin it and it's guaranteed, no silent substitution. Responses are the provider's, verbatim, and every call reports which model served it. The saving is in the rate, never in a quietly weaker model.

### Isn't a gateway in front of every call a single point of failure?

Kumo adds roughly 30–60ms p50 of overhead (under 120ms p99) on top of the model's own latency; public status history will be connected after the telemetry feed. If Kumo can't serve a request, it returns a contract-shaped 502, and you can always point your base URL straight back at the provider in one line.

### How do top-ups work?

You top up one prepaid balance — manually or with auto-recharge (set a threshold and a cap). Every call draws from it at the per-model rate you see before you spend. Unused balance remains spendable while your account stays active; token purchases are final and non-refundable.

### Refunds and support?

Token purchases are final and non-refundable, so the checkout asks you to confirm that before payment. Your tokens are not zeroed early: they remain spendable while your account stays active. Support is community plus email with next-business-day response, on a 99.9% uptime target; teams with committed volume get a dedicated contact.

### What are the rate limits and concurrency?

Today your throughput is bounded by your balance and per-key quota rather than a fixed per-second limit, so there's no separate rate-limit tier to manage. Need a guaranteed floor for a bursty workload? That's exactly what the committed-volume / team plan is for — talk to sales.

### Can your rates change? How much notice do I get?

Effective rates track the underlying provider list prices — when a lab cuts a price, yours drops too. If a rate ever has to rise we publish it in advance in the dashboard and by email, and committed-volume plans lock your rate for the term. You always see the exact per-model rate before you spend.

### What happens to my prepaid balance if something goes wrong with Kumo?

Token purchases are final and non-refundable, but purchased tokens remain spendable while your account stays active and are not zeroed early. For committed-volume agreements the unused portion is handled by the signed contract.

### Who's behind Kumo? Is it a real company?

Kumo is run by Kumo Cloud, Inc. — a real company, not a faceless reseller. A full team and company page is coming soon; until then, reach us directly at sales@openkumo.cloud and we'll answer any due-diligence question.

### What if something breaks at 3am?

If a provider error surfaces, you get a contract-shaped 502 and can switch providers directly. For Kumo-side issues, support is community plus email with a next-business-day target today; 24/7 paging and a status page with real history are landing in upcoming releases. Teams on committed volume get priority routing and a dedicated contact.

## Why Kumo vs alternatives

| Factor | Kumo | Going direct | Other gateways |
| --- | --- | --- | --- |
| Billing accounts to manage | One balance | One per vendor | One + markup |
| OpenAI-compatible API | Yes | Varies | Usually |
| Models behind one key | 9 active text models | 1 per account | Many |
| Fallback handling | Returns 502; you retry or switch provider | Build it yourself | Varies |
| Price vs list | 30–50% below list | List price | Varies |
| Usage & spend dashboard | Built-in, live | Per vendor | Basic |

## Security & trust

- **Your data isn't trained on** — Prompts and completions are never used to train models. Content is processed to serve your request and nothing more.
- **Prompt content isn't logged** — We store usage metadata for billing — not the body of your prompts. Bring your own retention policy.
- **Prompt bodies are not retained** — Kumo forwards prompts and completions to serve the call, then drops the content. We retain only the account, billing and usage metadata needed to operate the service.

## Links

- [Home](https://kumo.cloud/)
- [All models & pricing](https://kumo.cloud/models)
- [How it works](https://kumo.cloud/how-it-works)
- [Docs & quickstart](https://kumo.cloud/docs)
- [Blog](https://kumo.cloud/blog)
- [Kumo vs OpenRouter](https://kumo.cloud/vs-openrouter)
- [Security & Trust Center](https://kumo.cloud/security)
- [Status](https://kumo.cloud/status)
- [API base](https://api.kumo.cloud/v1)
- [llms.txt](https://kumo.cloud/llms.txt)