Ollama Cloud or an API gateway: what to choose for cloud models

Ollama is known as a program for running models on your own computer. Ollama Cloud continues the same line, but the models run on Ollama's servers, so you don't need powerful hardware at home. In parallel there is another path to cloud models: an API gateway with pay-per-token billing. Below we look at what Ollama Cloud is, how to get a key and call the API, how it differs from the Zukko gateway, and how to choose without guessing.

What Ollama Cloud is

According to the Ollama documentation, it is a cloud mode: you run models "in Ollama's cloud" from your apps or terminal without downloading anything. There are three ways to work:

• through the Ollama app or CLI after signing in to an account;
• through the REST API with a key;
• through OpenAI and Anthropic clients, which support part of the original APIs' features.

For the CLI the documentation also has agent launch commands, for example ollama launch claude, ollama launch codex, ollama launch opencode: Ollama configures such tools for its models. The set of cloud models changes; you can list them with curl https://ollama.com/api/tags or see the catalog on the Ollama site. The documentation also says models can have retirement dates, with warnings shown in the usage settings.

How to get a key and make a request

A key is created on the ollama.com settings page (the API keys section) and then put in an environment variable. A request per the documentation looks like this:

export OLLAMA_API_KEY="your_key"

curl https://ollama.com/api/chat \
  -H "Authorization: Bearer $OLLAMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma4:31b",
    "messages": [{"role": "user", "content": "Say hello in one sentence."}],
    "stream": false
  }'

You read the answer from message.content. The model name comes from the list that api/tags returns; the gemma4:31b example is taken from the documentation and may change over time. For OpenAI-compatible clients the documentation gives the address https://ollama.com/v1. Never leave the key in browser code or in a repository.

About subscriptions: the documentation says usage is visible in account settings and plans are compared on Ollama's pricing page. Plan terms and limits change, so we don't quote them here: see current information on the Ollama site.

How an API gateway works

The Zukko gateway is arranged differently. A single fetch_... key and one dollar balance give access to Claude, GPT, Gemini, DeepSeek, Qwen and GLM models at once. There is no subscription: you pay for tokens at the catalog price. Request formats are standard: OpenAI-compatible (https://api.zukko.pro/v1) and Anthropic-compatible (https://api.zukko.pro for clients that add /v1 themselves). Any model can be called in either format. Keys have a name, a daily and a monthly limit in $, and a list of allowed models.

Comparison

You can't compare "which is better" directly: the sets of models differ. Ollama Cloud makes sense if you already work in the Ollama ecosystem and want its models. A gateway makes sense if you need Claude or GPT models, payment for actual tokens and one key for different clients.

Counting on a concrete volume

To compare with a plan you need your monthly volume. Suppose 10M input tokens and 2M output tokens a month. Zukko prices as of October 18, 2026, $ per 1M tokens:

• glm-5.3-flash (input 0.04, output 0.04): 10 × 0.04 + 2 × 0.04 = $0.48;
• deepseek-v4.1-flash (0.013 / 0.038): 10 × 0.013 + 2 × 0.038 = $0.206;
• claude-sonnet-5-5 (0.4 / 2): 10 × 0.4 + 2 × 2 = $8.

Put these sums next to the price of the Ollama plan you see on their page, and next to the usage restrictions. At a small volume, pay-per-token may come out cheaper than a monthly plan; at a very large one you have to calculate separately. For agent sessions with long context, add a cache correction: for Claude the gateway sees the cache discount in the Anthropic format, and for GPT cache reads are counted in both. The general method is described in the article on the cost of an agent with cache↗.

One code path for both

Both services understand OpenAI-compatible requests, so you can write the code once and switch the address, key and model through settings:

import os
from openai import OpenAI

PROVIDERS = {
    "zukko": ("https://api.zukko.pro/v1", os.environ.get("ZUKKO_API_KEY"), "claude-sonnet-5-5"),
    "ollama": ("https://ollama.com/v1", os.environ.get("OLLAMA_API_KEY"), "gemma4:31b"),
}

base_url, key, model = PROVIDERS[os.environ.get("LLM_PROVIDER", "zukko")]
client = OpenAI(base_url=base_url, api_key=key)

resp = client.chat.completions.create(
    model=model,
    messages=[{"role": "user", "content": "Name three advantages of cloud models"}],
)
print(resp.choices[0].message.content)

The Ollama address and model name come from their documentation; verify current names with the api/tags request. Zukko serves its model list at GET https://api.zukko.pro/v1/models. More on the two formats: Anthropic or OpenAI↗. The Ollama documentation notes limits of the cloud API: for example, no stateful Responses mode and no built-in web search through /v1/responses, and clients support only a subset of the original APIs' features. If you rely on features specific to OpenAI or Anthropic, test them on the service you choose.

Access from Russia

For Zukko: access from Russia is through the gateway, with payment by SBP, MasterCard or crypto. As for Ollama Cloud, availability and payment methods must be checked on the service's own site: it depends on its terms, not on your settings. If payment or registration is difficult, that may be the deciding criterion.

How to choose

• You need models from the Ollama catalog and a familiar workflow with ollama in the terminal: try Ollama Cloud.
• You need Claude or GPT, pay-per-token billing, per-key limits and several clients on one balance: a gateway.
• You need both: the code above lets you keep both and switch by setting.
• Not sure: take one of your real tasks, run it on both options and compare the result and the cost. Drawing conclusions about quality from model lists is premature.

FAQ

Do I need Ollama Cloud if I run Ollama locally?

Not necessarily. A local run depends on your hardware and the model size. The cloud mode removes that limit, and a gateway adds models that can't be run locally at all.

Where do I get an Ollama API key?

In the account settings on ollama.com, in the keys section. Keep it in an environment variable and never publish it.

Can I use a Zukko key in Ollama?

No, these are separate services with separate keys. Code written with the OpenAI SDK can switch between them by changing base_url, the key and the model.

Does Zukko have a subscription?

No: a dollar balance and pay-per-token billing. Top-up from $1.

If you want to compare in practice, create a key in the dashboard↗ and run the example from the shared-code section: the comparison takes a few minutes.