The System One EventUtahGo To Event Page →

Clusters

Production capacity. Flat rate. No usage limits.

Rent a Midium-hosted cluster for your APIs and agents, or own the hardware — we build and install an on-site cluster for your office, colo, or air-gapped site.

01 · Cloud clusters

Rent Midium-hosted capacity.

Flat monthly rate. No token wallet. No usage limits. Your OpenAI-compatible APIs serve your team, your agents, and your production workloads.

02 · Owned clusters

Own the fleet. Host it with us or on site.

Go beyond a single Hub Mac with private Mac Studio clusters running the same Desktop OS at production scale. We can host that fleet on Midium infrastructure at a flat monthly rate (from $1,000/mo for a frontier-capable node), or build and install it wherever you need it: your office, your colo, your air-gapped site.

Desktop catalog

Models your cluster can run.

The full Midium Desktop catalog — agents, embeddings, vision, speech, and more. Design capacity around the models you actually need.

110 models

Midium Sub-Agent

Sub-Agent

  • Qwen3.5 4B

    Qwen3.5 · text-text

    Quant
    8bit
    Weights
    4 GB
    Context
    64K context
  • Gemma 4 E2B

    Preferred

    Gemma 4 · omni

    Quant
    QAT 4bit
    Weights
    5 GB
    Context
    128K context
  • Gemma 4 E2B

    Gemma 4 · omni

    Quant
    8bit
    Weights
    6 GB
    Context
    128K context

Pathfinder

  • Qwen3.5 2B

    Qwen3.5 · text-text

    Quant
    8bit
    Weights
    2 GB
    Context
    64K context
  • Qwen3 Embedding 0.6B

    Qwen3 · text-embedding

    Quant
    8bit
    Weights
    0.6 GB
    Context
    8K context

Indexing Model

  • Qwen3.5 2B

    Qwen3.5 · text-text

    Quant
    8bit
    Weights
    2 GB
    Context
    64K context

Embedding Model

  • Qwen3 Embedding 0.6B

    Qwen3 · text-embedding

    Quant
    8bit
    Weights
    1 GB
    Context
    32K context

Coding Agents

  • Mellum 2.1

    Mellum · text-text

    Quant
    6bit
    Weights
    9.9 GB
    Context
    131K context
  • Seed-Coder 8B

    Seed-Coder · text-text

    Quant
    6bit
    Weights
    7 GB
    Context
    33K context
  • Kimi-Dev 72B

    Kimi-Dev · text-text

    Quant
    4bit
    Weights
    38 GB
    Context
    131K context
  • Qwen3 Coder

    Qwen3 · text-text

    Quant
    8bit
    Weights
    30 GB
    Context
    256K context
  • Devstral 2

    Mistral · text-text

    Quant
    8bit
    Weights
    24 GB
    Context
    256K context
  • Qwen3 Coder Next

    Qwen3 · text-text

    Quant
    8bit
    Weights
    80 GB
    Context
    256K context

Midium Core Library

Lite Agents

  • Qwen3.5 9B

    Qwen3.5 · text-text

    Quant
    4bit
    Weights
    6 GB
    Context
    64K context
  • Gemma 4 E4B

    Preferred

    Gemma 4 · omni

    Quant
    QAT 4bit
    Weights
    7 GB
    Context
    128K context
  • Gemma 4 E4B

    Gemma 4 · omni

    Quant
    8bit
    Weights
    9 GB
    Context
    128K context
  • SmolLM3 3B

    SmolLM3 · text-text

    Quant
    4bit
    Weights
    2 GB
    Context
    128K context
  • LFM2.5 8B A1B

    LFM2.5 · text-text

    Quant
    4bit
    Weights
    4.9 GB
    Context
    128K context
  • LFM2.5 24B A2B

    LFM2.5 · text-text

    Quant
    4bit
    Weights
    14 GB
    Context
    131K context
  • Ministral 3 8B

    Mistral · text-text

    Quant
    4bit
    Weights
    5.6 GB
    Context
    64K context
  • Granite 4.0 H-Tiny

    Granite · text-text

    Quant
    4bit
    Weights
    3.9 GB
    Context
    64K context

Balanced Agents

  • Bonsai 27B

    Bonsai · omni

    Quant
    2bit
    Weights
    8.5 GB
    Context
    256K context
  • Gemma 4 12B

    Preferred

    Gemma 4 · omni

    Quant
    QAT 4bit
    Weights
    11 GB
    Context
    256K context
  • Gemma 4 12B

    Gemma 4 · omni

    Quant
    8bit
    Weights
    13 GB
    Context
    256K context
  • EXAONE 4.0 32B

    EXAONE · text-text

    Quant
    8bit
    Weights
    32 GB
    Context
    84K context
  • Gemma 4 26B A4B

    Preferred

    Gemma 4 · omni

    Quant
    QAT 4bit
    Weights
    14 GB
    Context
    256K context
  • Gemma 4 26B A4B

    Gemma 4 · omni

    Quant
    8bit
    Weights
    28 GB
    Context
    256K context
  • Gemma 4 31B

    Gemma 4 · omni

    Quant
    8bit
    Weights
    34 GB
    Context
    256K context
  • Gemma 4 31B

    Preferred

    Gemma 4 · omni

    Quant
    QAT 4bit
    Weights
    29 GB
    Context
    256K context
  • Qwen3.6 27B

    Qwen3.6 · omni

    Quant
    8bit
    Weights
    27 GB
    Context
    256K context
  • Qwen3.6 27B

    Preferred

    Qwen3.6 · omni

    Quant
    4bit
    Weights
    19 GB
    Context
    256K context
  • Qwen3.8 27B

    Qwen3.8 · omni

    Quant
    8bit
    Weights
    27 GB
    Context
    262K context
  • Qwen3.8 27B

    Qwen3.8 · omni

    Quant
    4bit
    Weights
    16 GB
    Context
    262K context
  • Qwen3.8 27B

    Preferred

    Qwen3.8 · omni

    Quant
    4bit
    Weights
    17 GB
    Context
    262K context
  • Bonsai 2 27B

    Qwen3.8 · omni

    Quant
    2bit
    Weights
    8.6 GB
    Context
    262K context
  • Qwen3.6 35B A3B

    Qwen3.6 · omni

    Quant
    8bit
    Weights
    35 GB
    Context
    256K context
  • Qwen3.6 35B A3B

    Preferred

    Qwen3.6 · omni

    Quant
    4bit
    Weights
    21 GB
    Context
    256K context
  • Flint

    Flint · text-text

    Quant
    4bit
    Weights
    23 GB
    Context
    262K context
  • OLMo 3 Think 32B

    OLMo 3 · text-text

    Quant
    4bit
    Weights
    17 GB
    Context
    66K context

Frontier Agents

  • Mistral Medium 3.5

    Mistral · omni

    Quant
    4bit
    Weights
    78 GB
    Context
    256K context
  • Nemotron 3 Super 120B A12B

    Nemotron 3 · text-text

    Quant
    4bit
    Weights
    63 GB
    Context
    262K context
  • INTELLECT-3

    INTELLECT · text-text

    Quant
    4bit
    Weights
    56 GB
    Context
    131K context
  • GLM-4.5V

    GLM-4.5V · image-text-text

    Quant
    4bit
    Weights
    58 GB
    Context
    66K context
  • Laguna S 2.1

    Preferred

    Laguna · laguna

    Quant
    4bit
    Weights
    60 GB
    Context
    1.0M context
  • Laguna S 2.1

    Laguna · laguna

    Quant
    6bit
    Weights
    92 GB
    Context
    1.0M context
  • Kolibri 1

    Preferred

    Kolibri · text-text

    Quant
    4bit
    Weights
    44 GB
    Context
    262K context
  • Kolibri 1

    Kolibri · text-text

    Quant
    8bit
    Weights
    83 GB
    Context
    262K context
  • Qwen3.8 Flash Next

    Qwen3.8 · omni

    Quant
    4bit
    Weights
    106 GB
    Context
    262K context
  • Ornith 1.5 122B

    Preferred

    Ornith · omni

    Quant
    4bit
    Weights
    73 GB
    Context
    262K context
  • GLM 5.3 Flash

    Preferred

    GLM · omni

    Quant
    oQ3.5e
    Weights
    161 GB
    Context
    1.0M context
  • GLM 5.3 Flash

    GLM · omni

    Quant
    oQ4e
    Weights
    186 GB
    Context
    1.0M context
  • GLM 5.3 Flash

    GLM · omni

    Quant
    oQ6e
    Weights
    264 GB
    Context
    1.0M context
  • GLM 5.3 Flash (REAP50 test)

    GLM · omni

    Quant
    mixed 4/8-bit
    Weights
    96 GB
    Context
    1.0M context

Embeddings

  • Qwen3 Embedding 8B

    Qwen3 · text-embedding

    Quant
    8bit
    Weights
    8 GB
    Context
    32K context
  • Qwen3 Embedding 4B

    Qwen3 · text-embedding

    Quant
    4bit
    Weights
    2.3 GB
    Context
    32K context
  • BGE-M3 (BAAI)

    BGE · text-embedding

    Quant
    8bit
    Weights
    2.5 GB
    Context
    512 context

Rerankers

  • BGE Reranker v2 m3 (BAAI)

    BGE · text-rerank

    Quant
    none
    Weights
    2.3 GB
    Context
    512 context
  • Qwen3 Reranker 0.6B

    Qwen · text-rerank

    Quant
    none
    Weights
    1.2 GB
    Context
    1K context
  • Qwen3 Reranker 4B

    Qwen · text-rerank

    Quant
    none
    Weights
    8 GB
    Context
    1K context
  • Qwen3 Reranker 8B

    Qwen · text-rerank

    Quant
    none
    Weights
    16 GB
    Context
    1K context

Image Generation

  • FLUX.1 schnell

    FLUX · text-image

    Quant
    —
    Weights
    31 GB
    Context
    80 context
  • FLUX.1 Kontext dev

    FLUX · text-image

    Quant
    —
    Weights
    31 GB
    Context
    80 context
  • FLUX.1 dev

    FLUX · text-image

    Quant
    —
    Weights
    31 GB
    Context
    80 context
  • FLUX.1 Krea dev

    FLUX · text-image

    Quant
    —
    Weights
    31 GB
    Context
    80 context
  • Z-Image Turbo

    Z-Image · text-image

    Quant
    —
    Weights
    31 GB
    Context
    512 context
  • Krea 2 Turbo

    Krea · text-image

    Quant
    —
    Weights
    33 GB
    Context
    1K context
  • FLUX.2 klein 4B

    FLUX.2 · text-image

    Quant
    —
    Weights
    15 GB
    Context
    512 context
  • Qwen-Image

    Qwen · text-image

    Quant
    —
    Weights
    54 GB
    Context
    512 context
  • Qwen-Image 2.1

    Qwen · text-image

    Quant
    —
    Weights
    31 GB
    Context
    512 context
  • Bonsai Image (2-bit, 4B)

    Bonsai · text-image

    Quant
    —
    Weights
    3.6 GB
    Context
    512 context
  • Qwen-Image-Edit

    Qwen · text-image

    Quant
    —
    Weights
    54 GB
    Context
    512 context
  • FLUX.1 Depth dev

    FLUX · text-image

    Quant
    —
    Weights
    41 GB
    Context
    80 context
  • SeedVR2 Upscaler (3B)

    SeedVR2 · text-image

    Quant
    —
    Weights
    14 GB
    Context
    512 context
  • SeedVR2 Upscaler (7B)

    SeedVR2 · text-image

    Quant
    —
    Weights
    34 GB
    Context
    512 context
  • Real-ESRGAN x4plus

    Real-ESRGAN · text-image

    Quant
    —
    Weights
    0.1 GB
    Context
    512 context
  • 4x RealWebPhoto v4 (DAT-2)

    DAT · text-image

    Quant
    —
    Weights
    0.1 GB
    Context
    512 context
  • 4x LSDIR (DAT)

    DAT · text-image

    Quant
    —
    Weights
    0.2 GB
    Context
    512 context
  • 4x NomosUni (DAT)

    DAT · text-image

    Quant
    —
    Weights
    0.2 GB
    Context
    512 context

Video Generation

  • Wan 2.2 TI2V 5B

    Wan 2.2 · image-text-video

    Quant
    8bit
    Weights
    17 GB
    Context
    512 context
  • Wan 2.2 I2V A14B

    Wan 2.2 · image-text-video

    Quant
    8bit
    Weights
    30 GB
    Context
    512 context
  • LTX-2.5

    LTX · image-text-video

    Quant
    distilled
    Weights
    67 GB
    Context
    512 context
  • LTX-2.5 Dev

    LTX · image-text-video

    Quant
    dev
    Weights
    80 GB
    Context
    512 context
  • LTX-2.3

    LTX · image-text-video

    Quant
    distilled
    Weights
    40 GB
    Context
    512 context
  • LTX-2.3 Dev

    LTX · image-text-video

    Quant
    dev
    Weights
    44 GB
    Context
    512 context

Speech

  • Whisper Large v3 Turbo

    OpenAI · audio-text

    Quant
    —
    Weights
    2 GB
    Context
    —

Vision Processing

Vision-Language Models

  • Qwen3-VL 8B

    Qwen3-VL · image-text-text

    Quant
    4bit
    Weights
    6 GB
    Context
    256K context
  • Qwen3-VL 30B A3B

    Qwen3-VL · image-text-text

    Quant
    4bit
    Weights
    17 GB
    Context
    256K context
  • Qwen3-VL 32B

    Qwen3-VL · image-text-text

    Quant
    4bit
    Weights
    19 GB
    Context
    256K context

Extra Models

Lite Agents

  • Gemma 4 E2B

    Preferred

    Gemma 4 · omni

    Quant
    QAT 4bit
    Weights
    5 GB
    Context
    128K context
  • Ministral 3

    Mistral · text-text

    Quant
    8bit
    Weights
    14 GB
    Context
    256K context
  • OLMoE

    OLMoE · text-text

    Quant
    8bit
    Weights
    7 GB
    Context
    4K context
  • Qwen3 4B Instruct 2507

    Qwen3 · text-text

    Quant
    4bit
    Weights
    2.3 GB
    Context
    64K context
  • Granite 3.3 2B

    Granite · text-text

    Quant
    8bit
    Weights
    4 GB
    Context
    64K context

Balanced Agents

  • GLM 4.7 Flash

    GLM · text-text

    Quant
    8bit
    Weights
    30 GB
    Context
    256K context
  • gpt-oss-20b (Fast)

    GPT-OSS · text-text

    Quant
    8bit
    Weights
    20 GB
    Context
    128K context
  • gpt-oss-120b

    GPT-OSS · text-text

    Quant
    4bit
    Weights
    60 GB
    Context
    128K context
  • Mistral Small 3.2

    Mistral · text-text

    Quant
    8bit
    Weights
    24 GB
    Context
    128K context
  • Qwen3.5 35B A3B

    Qwen3.5 · omni

    Quant
    8bit
    Weights
    35 GB
    Context
    256K context

Frontier Agents

  • GLM 4.5 Air

    GLM · text-text

    Quant
    8bit
    Weights
    106 GB
    Context
    128K context
  • Falcon H1

    Falcon · text-text

    Quant
    —
    Weights
    68 GB
    Context
    256K context
  • Solar Open

    Solar · text-text

    Quant
    8bit
    Weights
    100 GB
    Context
    128K context
  • Qwen3 Next

    Qwen3 · text-text

    Quant
    8bit
    Weights
    80 GB
    Context
    256K context

System

Decision Model

  • Midium Decider

    Decider · system_one

    Quant
    DWQ 4bit
    Weights
    1.1 GB
    Context
    33K context
  • Midium Decider 4B

    Decider · system_one

    Quant
    DWQ 4bit
    Weights
    2.4 GB
    Context
    33K context

Decision Models for Your Apps

  • d3-turbo

    vLLM Semantic Router · system_one

    Quant
    8bit
    Weights
    28 GB
    Context
    33K context
  • d3-flash

    vLLM Semantic Router · system_one

    Quant
    8bit
    Weights
    9.4 GB
    Context
    33K context
  • Perplexity Decider Flash v1.2

    Perplexity · system_one

    Quant
    8bit
    Weights
    5.1 GB
    Context
    33K context
  • Mammoth-1

    baselabs · system_one

    Quant
    8bit
    Weights
    29 GB
    Context
    33K context
  • Elephant-1

    baselabs · system_one

    Quant
    8bit
    Weights
    9.5 GB
    Context
    33K context
  • KnowLine 26B-A4B Gen2

    KnowLine · system_one

    Quant
    8bit
    Weights
    28 GB
    Context
    33K context

Throughput

What one Mac can actually serve

Measured on a single 128 GB M5 Max under sustained load, with simulated users based on real Midium usage.

Throughput

What one Mac can actually serve

Measured on a single 128 GB M5 Max under sustained load, with simulated users based on real Midium usage.

ModelCoding developers (Standard)Chat users (Standard)Background agents
Qwen3.6 35B4 (~6 at Relaxed)24 (~40 at Relaxed)2 at once, 238k tokens/h
Gemma 4 26B~5162 at once, 225k tokens/h
Ornith 1.5 122B212 at Relaxed4 at once, 134k tokens/h

Standard: coding 25 tok/s and first token within 5 s; chat 20 tok/s and 2 s. Relaxed: coding 15 tok/s and 10 s; chat 12 tok/s and 4 s.

  • Generation stays fast (50–220 tok/s per user). Reading long prompts sets the limit.
  • Qwen3.6 stays fast on long 128k-token coding sessions.
  • M5 Ultra and M3 Ultra: testing now. Estimates unpublished until measured.

Open-weight models

Host any open-weight models you want

Open weights keep getting better — so we don’t cap the story at a closed model. Size a cluster for the models you choose, on Macs Midium hosts or hardware at your site.

Monthly tokens

~1.1B

~1.0B in · ~95M out

Same tokens at closed-model list

~$2,960/mo

Closed frontier API list · $2 / $10 per 1M

128 GB Mac Studio

$600/mo flat

4.9× cheaper

A typical week of load on one node. The orange line is average load.Average loadMonTueWedThuFriSatSun
Modeled from 28 days of real customer usage patterns.
A chart with cost on the vertical axis and usage on the horizontal axis. Usage-based pricing climbs as usage grows. A cluster stays completely flat.

Cost vs usage

Usage-based pricing

Cluster

Usage-based pricing climbs with usage. A cluster stays flat.COSTUSAGE

Usage-based pricing climbs. A cluster stays flat.

Cluster

Your own AI cluster. Flat monthly rate.

Usage-based pricing climbs as traffic grows. A Midium cluster is a flat monthly rate for hardware we host, so production scale does not raise the bill.