What is this model?
DeepSeek V3.1 (V3.1-Terminus) is a large hybrid language model from DeepSeek for reasoning, coding,
tool-assisted workflows, and long-context technical work. On Ollama, the cloud tag is
deepseek-v3.1:671b-cloud — rated high usage, with 160K context, text input, tool support,
and thinking support. The same family offers a local deepseek-v3.1:671b build (~404 GB)
for teams that host inference on their own hardware.
One model supports both direct answers and extended reasoning by switching chat templates. That is useful when routine support tickets need fast replies but architecture reviews or policy analysis need deeper passes on the same deployment.
Why would I choose it?
Choose DeepSeek V3.1 Cloud when you need more reasoning depth than a small local model provides, but you are not ready to procure multi-GPU servers for a 671B-class checkpoint. It fits pilot programs, internal documentation Q&A, code review preparation, search-assisted agents, and long technical prompts — provided your data policy allows cloud inference.
For regulated material, confidential client data, or source code that must stay inside your network, use a private AI deployment with local inference instead. For a newer DeepSeek cloud option, see DeepSeek V3.2 Cloud.
What hardware should I plan for?
On a Terminal.Glass cloud-accelerated deployment, your server runs the one-server AI stack — not the 671B model weights. Plan a modest Linux host: roughly 4–8 GB RAM, two vCPUs, and 40 GB or more disk for Open WebUI, logs, and locally stored RAG indexes. A small VM, NUC, or mini-PC is typical. No GPU is required on the host.
If you run deepseek-v3.1:671b locally instead, size infrastructure for ~404 GB storage
and sustained multi-GPU inference. That is a separate deployment lane on customer-owned hardware, not
this cloud-accelerated workflow.
Ollama, Open WebUI, and OpenClaw
Ollama — Cloud models use the standard Ollama workflow. Sign in on the Terminal.Glass
host with ollama signin, then pull and run the cloud tag. Computation is offloaded to
Ollama Cloud per Ollama's cloud documentation.
ollama signin ollama run deepseek-v3.1:671b-cloud
Open WebUI — Serves as the team chat interface on your server. User accounts, chat history, and document indexes you build locally remain on hardware you control.
OpenClaw — Agent workflows can call this model for tool-calling prototypes, code agents, and structured analysis. Treat credentials, private keys, and proprietary logic as leaving your environment during cloud inference unless use has been explicitly approved.
Terminal.Glass fit and trade-offs
Terminal.Glass packages Ollama, Open WebUI, and OpenClaw into a one-server AI stack on your hardware or customer-owned cloud. For DeepSeek V3.1 Cloud, each prompt — including code excerpts, RAG passages, and tool context — is forwarded to Ollama Cloud; responses stream back through Open WebUI.
This is a cloud-accelerated private AI workflow, not fully local private AI deployment. Content processed by the model leaves your network. Usage is billed by Ollama at the high-usage tier; long 160K contexts and thinking-mode prompts increase cost. Expect network-dependent latency and less control over pinned model versions than a local checkpoint.
Deployment reference
| Tag | Recommended for |
|---|---|
| deepseek-v3.1:671b-cloud | Hybrid reasoning, tool use, code and search agents, long technical analysis when cloud processing is permitted |
| deepseek-v3.1:671b | Full on-prem inference when your hardware can host ~404 GB and sustain multi-GPU serving |
Confirm current tags and availability at ollama.com/library/deepseek-v3.1 before production. DeepSeek V3.1 weights are MIT-licensed per the Ollama model card; cloud access is separately governed by Ollama's terms.