HARDWARE

What needs several GPUs

Several GPUs are needed by flagship models, the ones comparable to the best cloud services. It is an expensive setup, and it makes sense only after a smaller model has been tested on your tasks and fell short.

13 familiesUpdated 22 Sep 2026Calculate video memory for your model size

What this means in practice

A 2-4 GPU serverLarge language models and sparse architectures. Usually rented rather than bought.
An 8 GPU serverThe largest open models and long video generation. Data-centre territory.
Hourly rentalThe sensible first step: rent a cluster while you test the idea instead of buying hardware for an unproven task.

What runs on it

Text 8

Text2026

Ripple-65B Preview

Ricursive Intelligence · USA

A dense (not MoE) model with 65B parameters, pretrained on 12 trillion tokens, with a context of up to 1 million tokens. It is built for programming and agents: by the authors' own test it scores 80.52 on Terminal-Bench 2.1 and beats much larger American open models (Reflection Beam 501B with 80.1, Meta Muse Spark Open 816B with 76.4, NVIDIA Nemotron 3 Ultra 550B with 56.4). The authors started from the Olmo 3 recipe and claim a 16 to 55 times gain in training efficiency on code. It was designed for NVIDIA graphics cards. BF16 and NVFP4 are published for the instruction-tuned version, plus a BF16 base. This is a preview release, and the figures are the authors' own, not independent.

  • an agent for programming in the terminal and a repository
  • long tasks with a context of hundreds of thousands of tokens
  • your own tuning on top of an open base model
Sizes
65B (dense)
Hardware
from: Cluster
Commercial use allowedDetails
TextRUGGUF2025–2026

MiniMax

MiniMax · China

Large MoE models with very long context (up to 1M tokens for Text-01 and M3). M3 is multimodal and understands images. In October the open weights of M3.1-Flash came out (about 428B, 23B active, up to 1M tokens of context, text, images and video): a model for programming, agents and computer use, sped up by sparse attention. MiniMax Community License: free for non-commercial use; commercial use needs the caption "Built with MiniMax M3.1-Flash" and a one-time notice to MiniMax, and a separate permission when the product earns $20 million a year or more; military use is prohibited. The benchmark figures come from MiniMax itself. Licenses differ greatly from version to version.

  • Analysis of large document archives in a single request
  • Agents with tools
  • Help for developers
Sizes
230B-A10B – 456B-A46B
Hardware
from: Cluster
Commercial use with conditionsDetails
Text2026

Beam

Reflection AI · USA

An open model from the American company Reflection with 501B parameters (23B active, a 1 million token context), text only, mainly English. Stated as economical in reasoning: according to the authors it needs about three times fewer tokens for thinking at comparable quality. Strong in programming, tool calling and science. Weights in BF16, FP8 and NVFP4. The benchmark figures are the authors' own.

  • an agent for solving tasks in a repository and a terminal
  • long tasks with tools and search (via MCP)
  • solving scientific and mathematical problems with reasoning
Sizes
501B-A23B
Hardware
from: Cluster
Commercial use allowedDetails
Text2026

Muse Spark Open

Meta Superintelligence Labs · USA

Meta's largest open model: 816B parameters, 42B active per token, a 1 million token context. It understands text, images, video and audio and answers in text. Built for programming and long agentic tasks. Trained on data in more than 100 languages. Access to the weights on Hugging Face is granted manually.

  • an agent for working with code and long tasks with tool calling
  • analyzing screenshots, charts, documents, video and recordings
  • generating training data for other models
Sizes
816B-A42B
Hardware
from: Cluster
Commercial use with conditionsDetails
Text2026

Atria Dawn

Shanghai Artificial Intelligence Laboratory · China

An agent model for long tasks: searching and gathering evidence, research, development, preparing reports and presentations, analyzing vulnerabilities in an authorized environment. A preview based on GLM-5.2, 256K-token context, text only. An FP8 version exists.

  • deep search and research on a topic with a verifiable result
  • developing programs and preparing reports and presentations
  • checking and fixing vulnerabilities in an authorized environment
Sizes
744B (MoE, based on GLM-5.2)
Hardware
from: Cluster
Commercial use allowedDetails
Text2026

Kolibri

Aleph Alpha · Germany

A European open reasoning model for German and English. A 78B mixture-of-experts with 3.5B active per token, a 262K-token context (tested up to a million) and tool calling.

  • answers over long documents and RAG
  • an agent that calls tools
  • extracting structured data from documents
Sizes
78B-A3.5B
Hardware
from: Cluster
Commercial use allowedDetails
TextGGUF2026

Inkling

Thinking Machines Lab · USA

Flagship open models from Mira Murati's lab: they take text, images and audio. Large MoE models that need several GPUs.

  • Flagship-level corporate assistant
  • Analysis of documents, images and audio
  • Programming help
Sizes
276B-A12B, 975B-A41B
Hardware
from: Cluster
Commercial use allowedDetails
Text2024–2025

Grok (открытые веса)

xAI · USA

xAI publishes the weights of previous Grok generations. The models are very large and need a GPU cluster, so in practice they are rarely run.

  • Research on large models
  • Assistant on your own infrastructure
  • Text generation and analysis
Sizes
314B (Grok-1), Grok-2 is larger
Hardware
from: Cluster
Commercial use with conditionsDetails

Video 3

Video2026

Pika H3 Flex

Pika Labs · USA

A Pika video model built on the open MiniMax H3 weights: it makes video with stereo sound at once in six steps and takes up to 12 reference materials (images, video, audio) that set the shot direction. Two adapters: FL2VA for text and the first and last frames, Ref2VA for reference materials. Locally it runs at 480p and 768p, while 1080p, 2K and 4K, audio stems and prompt enhancement are available only through the paid Pika API. It suits multi-shot stories, motion graphics and product films. The license inherits the MiniMax H3 terms and does not apply in the EU, the UK, South Korea and the USA, which has to be checked before use. Running it needs server-grade graphics cards.

  • ads and product films with ready sound
  • animation from reference images and video, multi-shot stories
  • motion graphics and title sequences
Sizes
33B (a fine-tune of MiniMax H3, adapters)
Hardware
from: Cluster
Commercial use with conditionsDetails
Video2026

MiniMax H3 (Hailuo)

MiniMax · China

Open weights of MiniMax's Hailuo video model. A large 33B model that makes video from text and images, but needs several server GPUs. The license does not apply in the European Union, the United Kingdom, South Korea and the USA. The October fine-tune of these weights is Pika H3 Flex.

  • Cinematic ad videos
  • Video from text and images
  • Complex scenes with motion
Sizes
33B + 32B encoder
Hardware
from: Cluster
Commercial use with conditionsDetails
Video2025

Step-Video

StepFun · China

A large 30B video model producing clips of up to 204 frames. Needs server hardware, but is open under MIT.

  • Video from a description
  • Animating images
Sizes
30B
Hardware
from: Cluster
Commercial use allowedDetails

Finance 1

FinanceNot maintained2024

Palmyra-Fin

Writer · USA

A large model for financial documents with a context window of about 131k tokens: it holds long reports whole. Not investment advice: decisions are made by a specialist.

  • Working with long annual reports and prospectuses
  • Summaries and digests of financial documents
  • Finding answers inside a large document pack
Sizes
70B (the model card states 72 billion parameters)
Hardware
from: Cluster
Commercial use with conditionsDetails

Fact-checking and judges 1

Fact-checking and judgesRU2024–2025

Nemotron Reward / GenRM

NVIDIA · USA

Large NVIDIA scorers for selecting and fine-tuning answers. The multilingual GenRM version lists Russian among its languages. The scorer itself makes mistakes and does not replace manual review on important tasks.

  • Choosing the best of several candidate answers
  • Preparing data to fine-tune your own model
  • Scoring assistant answers in Russian and other languages
Sizes
49B, 70B and 340B
Hardware
from: Cluster
Commercial use with conditionsDetails

Where people usually hit the wall

The usual mistake is to start with the flagship. In practice it goes the other way: a small model on one GPU, quality measured on your own examples, and only then a move up. A cluster adds more than hardware cost: networking, cooling, updates and someone on call.

By hardware

Need a model for your task?

An open model can run on your own server: data stays in-house, there is no per-request fee, and the model can be fine-tuned on your documents.

  1. SelectThe model and size for your task and hardware budget
  2. DeployOn your server or in a closed network, with an API
  3. Fine-tuneOn your data, or connect a knowledge base
  4. IntegrateInto your CRM, ERP, bot, website or team chat
Discuss deployment