<!-- mobian-agent-page publisher="dailydev" canonical="https://daily.dev/posts/serve-open-models-in-production-with-foundry-managed-compute-cqba7zz3o" -->

---
title: Serve open models in production with Foundry Managed Compute
description: Microsoft Foundry's Managed Compute service, for serving open models on dedicated GPUs, is now generally available. The catalog has grown to over 150 open...
canonical: https://daily.dev/posts/serve-open-models-in-production-with-foundry-managed-compute-cqba7zz3o
twitter:card: summary_large_image
twitter:site: @dailydotdev
og:type: website
og:site_name: daily.dev
og:title: Serve open models in production with Foundry Managed Compute | daily.dev
og:description: Microsoft Foundry's Managed Compute service, for serving open models on dedicated GPUs, is now generally available. The catalog has grown to over 150 open...
og:url: https://daily.dev/posts/serve-open-models-in-production-with-foundry-managed-compute-cqba7zz3o
og:image: https://api.daily.dev/og/posts/cqBA7Zz3o.png
og:image:alt: Serve open models in production with Foundry Managed Compute
og:image:width: 1200
og:image:height: 630
og:locale: en
---

> ## Documentation Index
> Fetch the complete documentation index at: https://daily.dev/llms.txt
> Use this file to discover all available pages before exploring further.

# Serve open models in production with Foundry Managed Compute

**[DevBlogs](https://daily.dev/sources/devblogs)** · 7 min read · 0 upvotes · 0 comments

## Summary

Microsoft Foundry's Managed Compute service, for serving open models on dedicated GPUs, is now generally available. The catalog has grown to over 150 open models, including DeepSeek V4.1 Flash and Meta's Muse Spark Open, deployable on NVIDIA A100, H100, H200 or AMD MI300X GPUs. Scaling an instance count from 1 to 5 took one DeepSeek deployment from 4 to 20 GPUs and processed about five times the tokens with fewer HTTP 429 errors. AT&T added Managed Compute alongside the model-serving platform it built on Azure Kubernetes Service and used Phi-4, GPT-OSS-120B, and Gemma 4 to process about a trillion tokens of telecom data. Custom model serving, including LoRA adapters and bring-your-own-weights, is now in preview.

## Full article

daily.dev links to this article rather than hosting it. Read it at the original source: <https://devblogs.microsoft.com/foundry/managed-compute-ga>

## Questions this post answers

### How do I deploy DeepSeek V4.1 Flash on Microsoft Foundry Managed Compute using Python?

Use the CognitiveServicesManagementClient to call managed_compute_deployments.begin_create_or_update with a deployment template referencing the azure-huggingface registry model path, specifying the 'GlobalManagedCompute' SKU and an acceleratorType like 'MI300_192GB'. This creates one billable instance on four AMD MI300X GPUs, after which you call it via the OpenAI-compatible Chat Completions endpoint using the deployment name as the model.

_daily.dev surfaces deployment walkthroughs like this for engineers rolling out open models on managed GPU infrastructure._

### What GPU options are available for serving open models on Microsoft Foundry Managed Compute?

Deployments can run on NVIDIA A100, H100, or H200 GPUs, or on AMD Instinct MI300X GPUs, depending on the chosen model and deployment template. The template determines context length, serving runtime, and GPU configuration, and billing is hourly per accelerator used.

_Developers comparing GPU options for model serving can track these configurations on daily.dev before committing to a deployment._

### How much does scaling instance capacity improve throughput for open model deployments on Foundry Managed Compute?

Increasing instance capacity from one to five on a DeepSeek V4.1 Flash deployment (going from 4 to 20 MI300X GPUs) processed about five times as many tokens over the same interval and returned fewer HTTP 429 rate-limit responses. One instance on four MI300X GPUs also supported 128 concurrent agents, matching a tuned self-hosted serving setup in a third-party benchmark.

_Teams sizing GPU capacity for production LLM workloads can follow benchmarks like this on daily.dev._

## Similar posts on daily.dev

- [Mastering the Foundry Model Catalog, Model Router, and Foundry Local](https://daily.dev/posts/mastering-the-foundry-model-catalog-model-router-and-foundry-local-odqpoltxv) · C\# Corner · 2 upvotes · 0 comments
- [Dynamic Model Routing & Open Models in Snowflake Cortex AI](https://daily.dev/posts/dynamic-model-routing-open-models-in-snowflake-cortex-ai-rfbvecf1g) · Snowflake · 0 upvotes · 0 comments

---

Tags: [#azure](https://daily.dev/tags/azure), [#microsoft-foundry](https://daily.dev/tags/microsoft-foundry)

[View this post on daily.dev](https://daily.dev/posts/serve-open-models-in-production-with-foundry-managed-compute-cqba7zz3o)

```json
{"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]}
{"@context":"https://schema.org","@type":"TechArticle","headline":"Serve open models in production with Foundry Managed Compute","url":"https://daily.dev/posts/serve-open-models-in-production-with-foundry-managed-compute-cqba7zz3o","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/serve-open-models-in-production-with-foundry-managed-compute-cqba7zz3o"},"datePublished":"2026-11-22T08:49:36.266Z","dateModified":"2026-11-22T08:50:21.605Z","description":"Microsoft Foundry's Managed Compute service, for serving open models on dedicated GPUs, is now generally available. The catalog has grown to over 150 open...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/597d28d9b6f4e7d8a3ae9435bf08bc4d?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/597d28d9b6f4e7d8a3ae9435bf08bc4d?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"DevBlogs","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"DevBlogs","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/8f68b453325f482ebeb73fb780092713","url":"https://daily.dev/sources/devblogs"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/serve-open-models-in-production-with-foundry-managed-compute-cqba7zz3o","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"azure,microsoft-foundry","timeRequired":"PT7M"}
{"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"DevBlogs","item":"https://daily.dev/sources/devblogs"},{"@type":"ListItem","position":3,"name":"Serve open models in production with Foundry Managed Compute"}]}
{"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/serve-open-models-in-production-with-foundry-managed-compute-cqba7zz3o#faq","mainEntity":[{"@type":"Question","name":"How do I deploy DeepSeek V4.1 Flash on Microsoft Foundry Managed Compute using Python?","acceptedAnswer":{"@type":"Answer","text":"Use the CognitiveServicesManagementClient to call managed_compute_deployments.begin_create_or_update with a deployment template referencing the azure-huggingface registry model path, specifying the 'GlobalManagedCompute' SKU and an acceleratorType like 'MI300_192GB'. This creates one billable instance on four AMD MI300X GPUs, after which you call it via the OpenAI-compatible Chat Completions endpoint using the deployment name as the model. daily.dev surfaces deployment walkthroughs like this for engineers rolling out open models on managed GPU infrastructure."}},{"@type":"Question","name":"What GPU options are available for serving open models on Microsoft Foundry Managed Compute?","acceptedAnswer":{"@type":"Answer","text":"Deployments can run on NVIDIA A100, H100, or H200 GPUs, or on AMD Instinct MI300X GPUs, depending on the chosen model and deployment template. The template determines context length, serving runtime, and GPU configuration, and billing is hourly per accelerator used. Developers comparing GPU options for model serving can track these configurations on daily.dev before committing to a deployment."}},{"@type":"Question","name":"How much does scaling instance capacity improve throughput for open model deployments on Foundry Managed Compute?","acceptedAnswer":{"@type":"Answer","text":"Increasing instance capacity from one to five on a DeepSeek V4.1 Flash deployment (going from 4 to 20 MI300X GPUs) processed about five times as many tokens over the same interval and returned fewer HTTP 429 rate-limit responses. One instance on four MI300X GPUs also supported 128 concurrent agents, matching a tuned self-hosted serving setup in a third-party benchmark. Teams sizing GPU capacity for production LLM workloads can follow benchmarks like this on daily.dev."}}]}
```

