Ollama Cloud changed its pricing on August 31, 2026. The old 5-hour session limits and weekly caps are gone. Every plan now includes a monthly pool of usage credits, spent per token at published rates: Free gets a small starter allowance, Pro ($20/month) includes $60 of usage, Max ($100/month) includes $300, and a new Team plan ($500/month) includes $1,000 shared across unlimited users.
If you read an older guide (including an earlier version of this one) that talks about “50x more usage” or 5-hour resets, that describes the legacy plans. Below is how the current plans work, what $60 of credits actually buys, and when each plan makes sense.
Would rather not pay per token at all? Running models on your own machine is still free and unlimited. Our Ollama hardware requirements guide shows what you need for each model size.
Ollama Cloud Plans at a Glance (2026)
| Free | Pro | Max | Team | |
|---|---|---|---|---|
| Price | $0 | $20/mo or $200/yr | $100/mo | $500/mo (intro) |
| Included usage | Small starter allowance | $60/mo | $300/mo | $1,000/mo shared |
| Models | Starter models (buy credits for all) | All, incl. larger pro models | All + early access to newest | All |
| Concurrent requests | 1 | 3 | 10 | 10 |
| Users | 1 | 1 | 1 | Unlimited |
| Usage resets | Monthly from signup date | Monthly on subscription date | Monthly on subscription date | Monthly |
The annual Pro plan works out to $16.67/month. Enterprise is custom-priced and adds per-model access controls and cost budgets per user and API key.
How the New Per-Token Credits Work
Every cloud request is billed in tokens at that model’s rate, split into input, cached input, and output. Your plan’s included credits are spent first. When they run out, you are not cut off: you can buy more credits and keep going at the same per-token price. There are no service fees on top.
- Included credits don’t roll over. Unused Pro credits are gone at your reset date; the pool refills to $60, not $60 plus leftovers.
- Cached input is much cheaper. On
deepseek-v4-pro, cached input is $0.044 per million tokens versus $1.32 uncached, which matters a lot for coding agents that resend the same context. - Off-peak pricing exists for some models. Outside 12:00–18:00 UTC on weekdays, and all weekend, models like
deepseek-v4-proanddeepseek-v4.1-flashcost half. - Free accounts can buy credits too. You don’t need a subscription to use the large models; you just pay per token from your first request.
What $60 (Pro) and $300 (Max) actually buy
This is the question the pricing page doesn’t answer directly, so here is a rough guide. It assumes 80% input and 20% output tokens with no cache hits, so real agent workloads with heavy caching will stretch further.
| Model | Input / output per 1M tokens | Pro ($60) | Max ($300) |
|---|---|---|---|
gpt-oss:20b | $0.07 / $0.30 | ~517M | ~2.6B |
gpt-oss:120b | $0.15 / $0.60 | ~250M | ~1.2B |
glm-5.3 | $1.40 / $4.40 | ~30M | ~150M |
deepseek-v4-pro | $1.32 / $3.96 | ~32M | ~162M |
kimi-k3 | $3.00 / $15.00 | ~11M | ~56M |
The takeaway: on a light model like gpt-oss:120b, Pro’s $60 is hundreds of millions of tokens and you will rarely hit it. On frontier-size models like kimi-k3 or deepseek-v4-pro, a busy coding agent can burn through $60 in a week or two. Rates are taken from Ollama’s pricing page and will change as models are added.
Concurrency: How Many Requests at Once
Free allows 1 concurrent request, Pro allows 3, and Max and Team allow 10. Requests beyond your limit are queued and processed as soon as a slot opens. The queue has a fixed size, though: if it fills up, new requests are rejected until a slot frees. If you run agents that fan out into parallel calls, this limit will bite before your credits do.
Already on Pro or Max? The Legacy Plan Question
If you subscribed before August 31, 2026, you stay on your old plan (with the 5-hour and weekly limits) as long as it keeps auto-renewing. You can switch to the new pricing in billing settings at any time, and your usage resets to the full monthly amount when you do.
One catch: changing your billing cycle (for example monthly to annual) moves you to the new pricing automatically. If you switch and regret it, Ollama says support can convert you back if you had an active subscription at launch.
Which is better for you depends on how you use it. If you ran heavy models in bursts and rarely hit the weekly cap, the legacy plan may give you more. If you kept hitting 5-hour limits mid-session, the new pool is friendlier because you can spend it whenever you like.
Available Cloud Models
Current cloud models include gpt-oss (20B and 120B), deepseek-v4-pro, deepseek-v4.1-flash, glm-5.3, kimi-k3, kimi-k2.7-code, minimax-m3, gemma4, mistral-large-3, and the nemotron-3 family. The full list with per-model prices is at ollama.com/search?c=cloud.
In the CLI you use the :cloud tag (for example glm-5.3:cloud) or the older -cloud suffix (gpt-oss:120b-cloud). Cloud models are never downloaded.
How to Use Ollama Cloud
From the CLI
ollama signin
ollama run gpt-oss:120b-cloud
# or launch a coding agent against a cloud model
ollama launch claudeYour local Ollama client forwards the request to Ollama’s cloud and streams the reply back. Local models keep working alongside it, so you can run llama3.2 locally for private work and a cloud model for the heavy lifting.
From the API (no local install needed)
export OLLAMA_API_KEY="your_api_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}'Create the key in your Ollama account settings. Ollama’s cloud also accepts OpenAI- and Anthropic-style clients for a subset of their APIs. For more request patterns, see our Ollama API guide.
Privacy and Hosting
Ollama says prompt and response data is never logged or trained on. Models run on NVIDIA Cloud Provider partners that must follow no-logging, zero-retention policies. Hosting is in the US and Europe, with Singapore used for a limited set of Qwen models.
Which Plan Should You Pick?
Stay on Free if you mainly run models locally and only want to try a cloud model now and then. If you need a big model occasionally, buying a small amount of credits is cheaper than a subscription.
Pick Pro ($20) if you would spend more than about $20/month in credits anyway. Pro gives you $60 of usage for $20, so it pays for itself the moment you use a third of it. It also lifts concurrency to 3.
Pick Max ($100) if you run agents all day on frontier-size models, or need 10 parallel requests. $300 of usage for $100 is the same 3x ratio as Pro, so the real reason to upgrade is concurrency and volume.
Pick Team ($500) if several people need access and you want one shared pool and one bill instead of individual subscriptions.
The self-hosted alternative: if per-token pricing doesn’t fit your usage, you can run the same open models on a GPU you rent by the hour and pay only for the time you use. RunPod is one option, with 24 GB cards from about $0.20/hour. That referral link gives new accounts a sign-up credit after a first $10 top-up; details in our affiliate disclosure.
Frequently Asked Questions
What are the Ollama Cloud free tier limits? The Free plan includes a small amount of monthly usage for a set of starter models, reset monthly from your signup date, with 1 concurrent request. To use the larger models you buy usage credits and pay the published per-token rate. There are no 5-hour or weekly limits anymore.
How much does Ollama Pro cost and what do you get? Pro is $20/month (or $200/year) and includes $60 of usage credits every month, access to the larger pro models, and 3 concurrent requests.
How much does Ollama Max cost? Max is $100/month and includes $300 of usage credits each month, 10 concurrent requests, and early access to new models.
Does Ollama Cloud still have 5-hour and weekly limits? Not on the new plans. Since August 31, 2026, Pro, Max and Team use a monthly pool of credits consumed per token. Subscribers who were on a legacy Pro or Max plan keep the old session and weekly limits until they switch.
What happens when I use up my monthly Ollama credits? You can keep going at the same per-token rate by buying usage credits. Included credits are used first, then your purchased balance. Unused included credits do not roll over.
Is there an Ollama Cloud team plan? Yes. Team is $500/month (introductory pricing) with $1,000 of shared monthly usage and unlimited users.
Does Ollama Cloud log my prompts? No. Ollama says prompt and response data is never logged or trained on, and its hosting partners must follow no-logging, zero-retention policies.
Final Verdict
The new pricing is simpler to reason about: you pay per token, the subscription buys credits at a 3x discount, and nothing locks you out for five hours. For most developers who use cloud models regularly, Pro is the obvious pick. Watch your model choice more than your plan, though: the same $60 is either effectively unlimited or gone in a week depending on whether you run gpt-oss or kimi-k3.
Last verified against Ollama’s pricing page and its August 31, 2026 pricing announcement on September 25, 2026.
Related Posts on DevToolHub
- Ollama vs LM Studio: Which One Should You Use
- Ollama Hardware Requirements: RAM, VRAM and GPU
- Ollama API: A Practical Guide with Examples
- Run Ollama on a DigitalOcean Droplet — Full Guide
- Best Open-Source LLMs for Local Running
- Docker vs Podman — Which Container Tool Should You Use?
- ArgoCD vs Flux — GitOps Tools Compared