awesome-ollama-models

Curated guide to Ollama local and cloud models, including plan limits, usage levels and model lifecycle status

Last verified: July 12, 2026

Ollama can run open models on your hardware or offload supported :cloud models to Ollama-managed infrastructure. Local execution does not consume Ollama Cloud usage. Cloud access is available through Free, Pro and Max plans.

Important

Ollama does not publish a stable model-by-model Free versus paid entitlement table. Model pages publish a relative cloud usage level instead. This guide separates official plan facts from practical model guidance so it does not present inferred access as a guarantee.

Contents

• Cloud plans
• Free versus paid guidance
• Current cloud models
• Quick decision guide
• Deprecations
• Setup
• API access
• Integrations
• Reliability guidance
• FAQ
• Contributing

Cloud plans

Plan	Price	Concurrent cloud models	Included cloud usage	Official positioning
Free	$0	1	Light	Chat, evaluation, coding and assistants with smaller models
Pro	$20/month or $200/year	3	50x Free	Larger models, coding automation and deep research
Max	$100/month	10	5x Pro	Sustained agents, concurrent workflows and extended large-model use

Ollama measures cloud usage primarily by GPU time rather than a fixed token or requests-per-minute allowance. Session limits reset every five hours and weekly limits reset every seven days. Running models on your own hardware remains unlimited from Ollama's cloud-plan perspective.

Always check the official pricing page for current terms.

Free versus paid guidance

Ollama model pages classify cloud consumption as low, medium, high or extra high usage. These are usage bands, not published plan entitlements.

Free plan starting points

Ollama currently labels two cloud models as low usage. Medium-usage models can also be useful for evaluation on Free, but they will consume the plan allowance faster.

Model	Usage	Context	Input	Good for
gpt-oss:20b-cloud	Low	128K	Text	Lightweight reasoning, coding and tool use
nemotron-3-nano:30b-cloud	Low	1M	Text	Efficient agent tasks and very long context
gpt-oss:120b-cloud	Medium	128K	Text	Stronger reasoning and coding
gemma4:cloud	Medium	256K	Text, image	Multimodal chat, coding and image analysis
qwen3.5:cloud	Medium	256K	Text, image	General, multilingual and multimodal work
deepseek-v4-flash:cloud	Medium	1M	Text	Long context, reasoning and coding
minimax-m2.7:cloud	Medium	200K	Text	Coding and agent workflows
minimax-m2.5:cloud	Medium	198K	Text	Coding and productivity
nemotron-3-super:cloud	Medium	256K	Text	Agentic and multilingual reasoning

Note

“Free plan starting point” does not mean unlimited or permanently guaranteed on Free. Confirm access while signed in and watch your account usage.

Paid-plan recommended

High and extra-high usage models are practical Pro or Max choices for repeated work. A Free account may allow limited evaluation, but this guide does not treat that as guaranteed access.

Model	Usage	Context	Input	Good for
glm-5.2:cloud	High	976K	Text	Long-horizon engineering and large repositories
glm-5.1:cloud	High	198K	Text	Agentic engineering and coding
mistral-large-3:675b-cloud	High	256K	Text, image	Multilingual and multimodal production tasks
minimax-m3:cloud	High	512K	Text, image	Coding, agents and long multimodal context
kimi-k2.7-code:cloud	High	256K	Text, image	Long-horizon coding and tool use
kimi-k2.6:cloud	High	256K	Text, image	Coding, design and agent orchestration
kimi-k2.5:cloud	High	256K	Text, image	General multimodal agent tasks
nemotron-3-ultra:cloud	High	256K	Text	Long-running, high-throughput agents
deepseek-v4-pro:cloud	Extra high	1M	Text	Frontier reasoning and difficult coding tasks

Current cloud models

The official cloud catalog changes frequently. The following active tags were verified on July 12, 2026.

Usage	Models
Low	gpt-oss:20b-cloud, nemotron-3-nano:30b-cloud
Medium	gpt-oss:120b-cloud, gemma4:cloud, gemma4:31b-cloud, qwen3.5:cloud, qwen3.5:397b-cloud, deepseek-v4-flash:cloud, minimax-m2.5:cloud, minimax-m2.7:cloud, nemotron-3-super:cloud
High	glm-5.1:cloud, glm-5.2:cloud, mistral-large-3:675b-cloud, minimax-m3:cloud, kimi-k2.5:cloud, kimi-k2.6:cloud, kimi-k2.7-code:cloud, nemotron-3-ultra:cloud
Extra high	deepseek-v4-pro:cloud

Models with a published retirement date are listed separately below instead of being recommended as current defaults.

Quick decision guide

Task	Lower-usage choice	Higher-capability choice
General work	qwen3.5:cloud	kimi-k2.5:cloud
Lightweight tasks	gpt-oss:20b-cloud	gpt-oss:120b-cloud
Coding assistant	minimax-m2.7:cloud	kimi-k2.7-code:cloud
Deep reasoning	deepseek-v4-flash:cloud	deepseek-v4-pro:cloud
Long documents	deepseek-v4-flash:cloud	glm-5.2:cloud
Images and screenshots	gemma4:cloud	minimax-m3:cloud
Multi-agent automation	nemotron-3-super:cloud	nemotron-3-ultra:cloud

Deprecations

Ollama occasionally retires cloud models. Retirement does not remove an independently available local model.

Already retired

Retired cloud model	Retirement date	Recommended replacement
kimi-k2-thinking:cloud	June 16, 2026	kimi-k2.6:cloud
kimi-k2:1t-cloud	June 16, 2026	kimi-k2.6:cloud
minimax-m2:cloud	June 16, 2026	minimax-m3:cloud
glm-4.6:cloud	June 16, 2026	glm-5.1:cloud
qwen3-next:80b-cloud	June 16, 2026	qwen3.5:cloud
qwen3-vl:235b-cloud	June 16, 2026	qwen3.5:cloud
qwen3-vl:235b-instruct-cloud	June 16, 2026	qwen3.5:cloud
cogito-2.1:671b-cloud	June 16, 2026	deepseek-v4-flash:cloud
rnj-1:8b-cloud	June 30, 2026	None listed

Scheduled for retirement July 15, 2026

Do not use these tags for new integrations.

Retiring cloud model	Recommended replacement
deepseek-v3.1:671b-cloud	deepseek-v4-flash:cloud
deepseek-v3.2:cloud	deepseek-v4-flash:cloud
devstral-2:123b-cloud	mistral-large-3:675b-cloud
devstral-small-2:24b-cloud	None listed
ministral-3:3b-cloud	None listed
ministral-3:8b-cloud	None listed
ministral-3:14b-cloud	None listed
gemini-3-flash-preview:cloud	minimax-m3:cloud
gemma3:4b-cloud, gemma3:12b-cloud, gemma3:27b-cloud	gemma4:31b-cloud
glm-4.7:cloud, glm-5:cloud	glm-5.2:cloud
minimax-m2.1:cloud	minimax-m3:cloud
qwen3-coder-next:cloud, qwen3-coder:480b-cloud	qwen3.5:397b-cloud

Check the official cloud deprecations before merging model changes.

Setup

Install Ollama, then sign in for cloud access:

ollama signin

Run a cloud model:

ollama run qwen3.5:cloud "Explain Kubernetes admission control"

Run a local model without consuming cloud usage:

ollama run qwen3.5:9b "Format this JSON"

API access

Use a cloud tag through the local Ollama endpoint:

curl --fail-with-body --silent --show-error \
  --connect-timeout 10 \
  --max-time 120 \
  http://localhost:11434/api/chat \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "qwen3.5:cloud",
    "messages": [{"role": "user", "content": "Explain Kubernetes admission control"}],
    "stream": false
  }'

The direct Ollama Cloud API uses the exact model names returned by /api/tags. These names do not use the local :cloud suffix and may include a size tag. Create an API key in your Ollama account, then:

curl --fail-with-body --silent --show-error \
  --connect-timeout 10 \
  --max-time 120 \
  https://ollama.com/api/chat \
  -H "Authorization: Bearer ${OLLAMA_API_KEY}" \
  -H 'Content-Type: application/json' \
  -d '{
    "model": "gpt-oss:120b",
    "messages": [{"role": "user", "content": "Explain Kubernetes admission control"}],
    "stream": false
  }'

List models currently exposed by the direct cloud API:

curl --fail-with-body --silent --show-error https://ollama.com/api/tags

See the official cloud API documentation for current authentication and tag conventions.

Integrations

Cloud models use the same Ollama interfaces as local models.

Tool	Purpose	Link
Open WebUI	Browser chat interface	github.com/open-webui/open-webui
Continue	VS Code and JetBrains assistant	continue.dev
LangChain	LLM application framework	python.langchain.com
LlamaIndex	RAG and data framework	llamaindex.ai
Aider	Terminal coding assistant	aider.chat
ollama-model-router	Route prompts among Ollama models	github.com/RedBeret/ollama-model-router

Reliability guidance

• Do not hard-code a cloud model without a fallback.
• Prefer active models over tags with a published retirement date.
• Treat HTTP 401 and 403 as authentication or access failures.
• Treat HTTP 429 as a rate-limit response and honor Retry-After when present.
• Retry transient 5xx failures with capped exponential backoff.
• Set connection and response timeouts.
• Keep a local fallback when privacy, availability or cost is critical.
• Do not log API keys or sensitive prompts.

Example fallback order:

qwen3.5:cloud
  -> gpt-oss:20b-cloud
  -> qwen3.5:9b (local)

FAQ

Are Ollama cloud models free?

Ollama has a Free cloud plan with light usage. Pro and Max provide more usage and concurrency. Pro also advertises access to larger, more powerful cloud models. It is no longer accurate to call every cloud model universally free.

Which exact models are on the Free tier?

Ollama does not publish a stable per-model entitlement matrix. Start with low-usage models, use medium-usage models cautiously and confirm access while signed in. The usage bands in this guide describe relative compute consumption, not guaranteed plan eligibility.

Are local models free?

Running public models on your own hardware does not consume Ollama Cloud usage. Your hardware, electricity and infrastructure costs still apply.

What are the rate limits?

Ollama publishes session and weekly usage limits rather than a fixed requests-per-minute allowance. Concurrency is one cloud model on Free, three on Pro and ten on Max.

Can cloud models run offline?

No. Use a local model tag for offline operation.

Sources

• Ollama pricing
• Ollama Cloud documentation
• Ollama cloud model catalog
• Individual Ollama model pages linked above

Contributing

Before adding or changing a cloud model:

1. Confirm it appears in the official cloud catalog.
2. Confirm the exact tag, context, input types and usage level on its model page.
3. Check the cloud deprecations list.
4. Avoid unsupported benchmark, latency, price or rate-limit claims.
5. Include the verification date in the pull request.

Suggested row format:

| model-tag | usage | context | input | good for |

License

MIT