Standard Compute

Open-source models for business, through the same API as Claude and GPT

Standard Compute gives your company leading open models, including GLM, DeepSeek, MiniMax and Kimi, alongside Claude, GPT and Gemini through one API, and you decide which applications may use which models.

Create a company workspaceor write to contact@standardcompute.com

Short answers

Which open-source models can a business use through Standard Compute?
GLM 5.3, GLM 5.3 Flash and GLM 5.2 from Z.ai, DeepSeek V4 Pro and V4.1 Flash from DeepSeek, MiniMax M3 and M3.1 Flash from MiniMax, Kimi K3 from Moonshot AI, Qwen 3.8 27B from Alibaba and MiMo V2.6 from Xiaomi, all through the same API and account as Claude, GPT and Gemini.
Which open models are served by European providers?
GLM, DeepSeek and MiniMax are also served by European providers for Europe-restricted workspaces. Set a Europe rule for your company, a team or a single key, and every request, including retries and fallbacks, is served only by approved European inference endpoints.
Do we need our own GPUs?
No. Your applications call one API at https://api.stdcmpt.com/v1, or https://eu.stdcmpt.com/v1 for our EU gateway, and the models are served by inference providers you can approve one by one.
How do open models lower the bill?
A single API gives you frontier and open-source models, and it routes each step to the cheapest model that can do it and the cheapest provider fast enough to serve it. Routine steps go to open models and the hard reasoning step goes to a frontier model.
Can we keep open models to the applications we choose?
Yes. Decide which models each key can use, so every app and every team member gets exactly the models you approve.

The open models in the catalog

Every model below is available today through the same endpoint, key and invoice as the frontier models. The model names link to each developer's official model card, where the weights and the licence are published.

Open-weight models available through Standard Compute
ModelDeveloperLicence on the model cardEuropean providers
GLM 5.3Developer: Z.aiLicence on the model card: GLM 5.3 licenceEuropean providers: TensorX, Inceptron, Mistral
GLM 5.3 FlashDeveloper: Z.aiLicence on the model card: MITEuropean providers: TensorX, Inceptron
GLM 5.2Developer: Z.aiLicence on the model card: MITEuropean providers: TensorX, Inceptron, Mistral
DeepSeek V4 ProDeveloper: DeepSeekLicence on the model card: MITEuropean providers: TensorX (V4 Pro 0813)
DeepSeek V4 FlashDeveloper: DeepSeekLicence on the model card: MITEuropean providers: TensorX, Inceptron (V4 Flash 0731)
DeepSeek V4.1 FlashDeveloper: DeepSeekLicence on the model card: MITEuropean providers: TensorX
MiniMax M3Developer: MiniMaxLicence on the model card: MiniMax community licenceEuropean providers: TensorX
MiniMax M3.1 FlashDeveloper: MiniMaxLicence on the model card: MiniMax community licenceEuropean providers: TensorX
Kimi K3Developer: Moonshot AILicence on the model card: Kimi K3 licenceEuropean providers: None; served globally
Qwen 3.8 27BDeveloper: Alibaba (Qwen team)Licence on the model card: Apache 2.0European providers: None; served globally
MiMo V2.6Developer: XiaomiLicence on the model card: MITEuropean providers: None; served globally

Licences as published on each official model card on Hugging Face, checked 2026-11-07. European providers from the endpoint list the gateway enforces, reviewed 2026-10-27. Live per-token prices for every model are on /models.

Open models on the applications you choose, frontier models for the hard steps

Most companies do not need one model for everything. A support classifier, a document extractor or an internal search tool runs well on an open model. A legal review or a multi-step coding agent gets a frontier model for the step that needs it.

Every person and every app has its own key, and each key carries its own list of approved models. Give the extraction service GLM 5.3 and DeepSeek V4.1 Flash, give the engineering team Claude Opus 5.5 as well, and a request for a model outside a key's list is rejected.

Within those lists, each model is served by the cheapest provider that meets the latency and tokens-per-second bar for the job.

curl https://api.stdcmpt.com/v1/chat/completions \
  -H "Authorization: Bearer $EXTRACTION_SERVICE_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "z-ai/glm-5.3",
    "messages": [{"role": "user", "content": "Extract the invoice number, date and total."}]
  }'
A request from an application key that is approved for GLM 5.3. Switching model is a change to one field.

Governance for open models

Choose exactly which inference providers may handle your data, for the whole company, a team or a single key. Lower levels can only narrow the rules, new providers are never added without an admin's approval, and fallbacks can never route around your list.

Turn on zero data retention for your company, a team or a key, and every request is routed only to inference endpoints that commit to keeping no copy of your data. Standard Compute never trains on your prompts or outputs. With a region rule or zero retention turned on, every request goes only to providers that commit not to train on it.

Every admin change is recorded with before and after values, and you can export a record of every request showing the model, the provider that served it and the region rule in force, kept for 12 months.

Three ways to run open models in a company

Ways to run open models
ApproachWhat your team managesFits when
Your own GPUsWhat your team manages: Hardware, serving software, scaling, patching and capacity planningFits when: You run one model at steady high volume and have a platform team for it
One inference provider per modelWhat your team manages: A contract, a key, a DPA review and a bill for each providerFits when: One model family covers all your work
Standard ComputeWhat your team manages: One company account, one DPA, rules per team and keyFits when: Several teams use open and frontier models and security wants one set of controls

Questions from IT, security and procurement

Which open-source LLMs are good enough for business use, and where do we run them without hosting GPUs?
Their developers build GLM 5.3 for complex coding and long-horizon tasks, DeepSeek V4.1 Flash for input-heavy agentic work and MiniMax M3 for coding and long-horizon agent work. Through Standard Compute you call them on one API, served by inference providers you approve, with no GPUs to run.
We want open-weight models hosted in Europe with no data retention. Who does that?
Standard Compute does. GLM, DeepSeek and MiniMax are also served by European providers for Europe-restricted workspaces, and turning on zero data retention routes every request only to inference endpoints that commit to keeping no copy of your data.
How do we route easy requests to cheap open models and hard ones to Claude or GPT?
Approve both kinds of model on the key and let the router decide per step. Routine steps land on open models, and the hard step goes to Claude or GPT.
Is it cheaper to use open-source models through an API than GPT or Claude?
Per token, yes. DeepSeek V4.1 Flash lists at $0.30 and GLM 5.3 at $1.40 per million input tokens, against $4 for Claude Opus 5.5; live prices for every model are on /models. Every person and app gets a fixed monthly budget that never overruns, so the saving shows up as more work done inside the same budget.
Does using DeepSeek send our data to China?
Set a Europe rule and every DeepSeek request, including retries and fallbacks, is served only by approved European inference endpoints. A provider allow-list lets you name exactly which providers may serve any model.
Are these models open source or open weight?
They are open-weight models: the developers publish the weights under the licence on each model card. DeepSeek V4, GLM 5.2 and MiMo V2.6 use MIT, Qwen 3.8 27B uses Apache 2.0, and GLM 5.3, MiniMax M3 and Kimi K3 use their developer's own licence.
Do you offer an SLA and invoices?
Enterprise includes a 99.9% monthly uptime SLA with service credits. Plans are flat and monthly per person or app, with Stripe invoices that carry your company name and VAT number.
Can IT restrict which teams may use open models?
Yes. Decide which models each key can use, and set provider rules for the company, a team or a single key. A lower level can narrow a rule set above it and never loosen it.

Related pages

Create your company, approve the models each team may use and connect your identity provider from the dashboard.

Create a company workspace