API tokens at
subscription prices
Every leading model at subscription prices — Claude, GPT, Gemini, Grok, DeepSeek and more through one API. One key, one balance, no substitutes or distillates: you always get exactly the model you asked for.
Models in the pricing in force
Almost every availability check came back clean
Today October 2, 2026
Uptime
Stable routing
Kumo runs on a stable routing system, which keeps uptime high across every kind of workload
Measured from the platform's own availability record Full status →
Three steps to your first request
Sign up, top up once, grab a key — and you're calling supported GPT, Claude and Gemini text models from one balance. No sales call, no contract, no per-seat fees.
Sign up
Email and a password — no card, VPN or foreign number. Ready in a minute
Top up once
Russian card, SBP or crypto. The promo code's bonus lands with your first top-up
Get your API key
Create a key and start shipping. Fully OpenAI-compatible — change one base URL
Generate a key to send your first request
One API gateway to every frontier model
GPT, Claude, Gemini, Grok, DeepSeek at subscription prices — the originals, no substitutes or distillates. Switch models by changing one identifier in the request
One API gateway to every frontier model
GPT, Claude, Gemini, Grok, DeepSeek at subscription prices — the originals, no substitutes or distillates. Switch models by changing one identifier in the request
The catalogue in figures. This many models and vendors behind one key right now — counted off the price list in force, not off a promise
The request goes to Anthropic's own API; the model you named answersThe request goes to the vendor's own API. No distillates, no "compatible" or substitute models: the answer comes from the very model you named.
Strictly original models. Start with a small key and verify it yourself before trusting it with production.
The request passes through to the vendor's model named in it and backKumo does not read the bodies of your prompts and answers: the request passes through to the vendor's model named in it and comes back.
Complete anonymity. We do not look into your requests and keep no logs of them — only the metadata your bill is made of
Pay by QR from any Russian bank's app; the money lands on the same balancePay by QR code from any Russian bank's app. No card and no details to type; the money lands on the same single balance every model is charged to.
Flexible pricing, easy top-up from Russia. Packages or pay-as-you-go — whichever fits you. Top up by card or SBP, no VPN or foreign cards
Package
Steady load
- Developers and teams calling models every day
- Services, agents and pipelines with constant traffic
- Large volumes where the lowest per-token rate matters
API balance
Light or uneven load
- Day-to-day work and small projects
- Trying different models on separate keys
- One-off tasks, prototypes and experiments
A package or the API balance. Steady load — a package at the lower rate; light or uneven load — the balance with no commitment. The names lead to the pricing page
export ANTHROPIC_BASE_URL="https://api.kumorouter.com"
export ANTHROPIC_AUTH_TOKEN="$KUMO_API_KEY"
export ANTHROPIC_MODEL="anthropic/claude-opus-4-8"
claudeEditors and CLIs. Claude Code, Codex CLI, Cursor, Cline, Continue, Zed — a config file each, and nothing else
- 1
The key
kumo_sk_•••••••••••••••• - 2
The address
- https://api.openai.com/v1+ https://api.kumorouter.com/v1 - 3
The first call
anthropic/claude-opus-4-8200 OK
Three steps, and not a line more. Onboarding hands you the key, the gateway address and the setup prompt for your tool — pasting it into the config is what is left
OpenAI-compatible API
The same official SDK and the same operations: chat completions, responses, embeddings, images
Native Anthropic protocol
Clients written for Anthropic Messages connect on their own base URL, unchanged
What it runs on. One gateway address for both protocols — a client connects on the one it was written for and rewrites nothing
Below list on every model. Flagship Claude, GPT and Gemini or low-cost DeepSeek and Qwen — every one of them the vendor's own model, and every one billed under the provider's list price, not just the cheap ones.
Packages deepen the discount. Prepay a volume and the effective rate drops further; pay-as-you-go stays there for the months you cannot predict
The rate is published, not a surprise. Per-1M input and output rates are on the price page and in your dashboard — the exact cost of a call before you make it
Several upstreams per model. Where the catalog lists more than one provider endpoint for a model, the request goes to the one that answers
Switch models without a rewrite. The model is one string in the request. Compare answers, move a workload, mix cheap and flagship — the client never changes
Errors you can act on. A provider failure comes back as a contract-shaped 502, not a hang — retry or switch providers from your own code
Keys with a scope. One per project, agent or machine — issued and revoked in the console, each with its own caps
Month-to-date spend per project. Real per-project cost, straight from the keys inside it — the number you re-bill or put in COGS
A runaway loop can't drain the balance. Daily and monthly caps on every key, live usage and alerts — the blast radius is one key, not the account
One balance across every vendor. Anthropic, OpenAI and Google spend on a single balance with a report finance can reconcile — no per-vendor accounts, no foreign cards
Strictly original models. The vendor's own model answers, every time. Start with a small key and verify it yourself before trusting it with production.
Every model in one catalog. Live per-1M input and output rates for the whole list; a new model is a new identifier, not a new integration
Latest from the blog
Product updates, honest comparisons and user stories — written by the team that runs the gateway
Tell us what you need
A line about your project or question — we’ll route it to the right person
- sales@kumorouter.com Sales & B2B
- support@kumorouter.com Support
- @kumo_api News on Telegram
Indie developer
A solo builder shipping a side project after hours
Runs GPT-5.6 Terra and Claude behind one base URL on a small top-up that lasts
One key, one balance — no five-vendor juggling. Just ship