@meta
  v: 1
  route: /model/mixtral-8x7b
  generated: 2026-08-01T10:49:02.913Z

@intent
  purpose:    Show what hardware can run Mixtral 8x7B.
  audience:   ai-engineer, self-hoster, model-evaluator
  capability: inspect_model_specs, find_compatible_gpus, compare_similar_models, open_huggingface

@state
  slug: mixtral-8x7b
  name: Mixtral 8x7B
  hf_repo: mistralai/Mixtral-8x7B-Instruct-v0.1
  params_b: 46.7
  active_params_b: 12.9
  family: Mistral
  type: moe
  context_k: 32
  fp16_gb: 93.4
  summary: "The MoE that proved sparse models work for inference. 8 experts, 2 active per token."
  smallest_fitting_gpu_q4: a6000
  smallest_fitting_gpu_fp16: m4-max-128

@actions
  - id: open_huggingface
    method: GET
    href: https://huggingface.co/mistralai/Mixtral-8x7B-Instruct-v0.1
  - id: view_compatible_gpus
    method: GET
    href: /model/mixtral-8x7b
  - id: compare_similar
    method: GET
    href: /model/qwen-3-6-35b-a3b
  - id: view_calculator
    method: GET
    href: /#calculator

@context
  > The MoE that proved sparse models work for inference. 8 experts, 2 active per token.

@nav
  self:      /model/mixtral-8x7b
  parents:   [/]
  peers:     [/model/qwen-3-6-35b-a3b, /model/yi-34b, /model/qwen-2-5-32b, /model/qwen-2-5-coder-32b]
  drilldown: /the-math