Deploy Mistral Medium 3.5 128B on GPU Cloud (Full Tutorial)

About this lesson

Deploy Mistral Medium 3.5, Mistral AI's new open-weight 128B flagship model on Hyperstack GPU cloud using vLLM. This step-by-step tutorial covers VM setup, vLLM nightly installation, serving the model across 8x H100 GPUs and using its reasoning mode and function-calling features. Mistral Medium 3.5 is the first merged Mistral model: a single 128B dense checkpoint that replaces three separate products (Mistral Medium 3.1, Magistral, and Devstral 2) with unified instruction following, coding, vision, and configurable reasoning in one deployment. It supports a 256K token context window and scores 77.6% on SWE-Bench Verified. In this tutorial, you'll learn: What makes Mistral Medium 3.5 different from previous Mistral models How to provision an 8x H100-80GB PCIe (NVLink) VM on Hyperstack How to install vLLM nightly (required for Mistral 3.5 architecture support) How to launch the OpenAI-compatible inference server with tensor parallelism How to use reasoning_effort="high" vs "none" per request How to enable streaming and native function calling How to use EAGLE speculative decoding to reduce latency Full tutorial on Hyperstack Blog: https://bit.ly/3RMSfkG Get started on Hyperstack: https://bit.ly/43S7UBK Like and subscribe for GPU cloud and open-source LLM tutorials! #MistralMedium #Mistral35 #LLMDeployment #vLLM #Hyperstack #H100 #OpenSourceLLM #AITutorial

Original Description