Show HN: Tokenstead, find AI models for your hardware

cdnsteve1 pts0 comments

Tokenstead - Find AI Models for Your Hardware

●LIVE<br>34 MODELS<br>56 HARDWARE<br>22 ADOPTERS TRACKED<br>38 UNIFIED-MEM DEVICES<br>HOMESTEAD YOUR AI

◆ Homestead your AI

Find the right model for your hardware

Run AI on hardware you own, so no vendor or government order can switch off the model you depend on. Pick your rig and see which open models fit - with honest speed estimates, cloud-pricing comparisons, and a source-cited tracker of who's running what.

01 - Select your hardware

click to compare

AMD Ryzen AI Halo 128GB

128GB UNIFIED

Framework Desktop 128GB

128GB UNIFIED

GMKtec EVO-X2 128GB

128GB UNIFIED

Mac Mini M4 16GB

16GB UNIFIED

Mac Mini M4 Pro 24GB

24GB UNIFIED

Mac Mini M4 Pro 48GB

48GB UNIFIED

Mac Studio M4 Max 36GB

36GB UNIFIED

Mac Studio M4 Max 64GB

64GB UNIFIED

Mac Studio M4 Max 96GB

96GB UNIFIED

or enter custom specs

On a budget? Compare own vs rent →

Count tokens & estimate API cost →

Unified (Mac/Jetson)<br>VRAM (GPU card)<br>System RAM

02 - Save your rig (free)

Sign in with GitHub to save your hardware. New here? We'll guide you through picking your rig - Mac, multi-GPU (up to 8x), or custom specs - then show you exactly which models you can run locally.

Sign in with GitHub - free

Top models by quality

Browse all models

01

DeepSeek V4 Pro<br>MoE

1600.0B params - 1000k ctx

89<br>GENERAL

02

GLM 5.2<br>MoE

744.0B params - 1000k ctx

88<br>GENERAL

03

Qwen3.6 35B A3B<br>MoE

35.0B params - 262k ctx

87<br>GENERAL

04

DeepSeek V3 0324<br>MoE

671.0B params - 128k ctx

87<br>GENERAL

05

DeepSeek V4 Flash<br>MoE

284.0B params - 1000k ctx

86<br>GENERAL

06

LongCat 2.0<br>MoE

1600.0B params - 1000k ctx

86<br>GENERAL

Who's running what

See all 22 →

Microsoft<br>runs<br>MAI<br>reported<br>Microsoft

Bloomberg: Microsoft replacing OpenAI/Anthropic models with in-house MAI models in Excel and Outlook to cut inference spend; a tuned MAI variant claims GPT-5.4 parity at up to 10x efficiency. Proprietary (not open-weight), but the same frontier-API cost pressure.

Smartly<br>runs<br>Llama 3.1 8B<br>confirmed<br>Meta

Self-hosted Llama 3.1 8B on Kubernetes automates support-ticket creation and resolution drafts for the ad-tech platform; 80% less time to create tickets.

Caisse des Depots<br>runs<br>Mistral Medium 3.5<br>confirmed<br>Mistral AI

Mistral Medium 3.5 (128B) for up to 100k French public-sector agents under a 4-year, EUR 140M framework; on-prem SecNumCloud option for sovereignty.

Capgemini<br>runs<br>Codestral<br>confirmed<br>Mistral AI

Self-hosted Codestral in its RAISE/SovBox coding assistant for regulated aerospace, defense and public-sector clients; code-completion accuracy 50% -> 90%.

SAP<br>runs<br>Codestral<br>confirmed<br>Mistral AI

Self-hosted, 100% European Codestral powers multilingual SBB rail support bots (1k -> 30k employees, 80% fewer repetitive queries) and an accounting accruals agent.

Statuses:<br>confirmed official source<br>reported credible third-party, not officially confirmed<br>testing evaluating, not deployed.

Curated and source-cited, not a scraper. Built a rig worth sharing?<br>Browse shared builds →.

34

Models tracked

56

Hardware configs

22

Adopters tracked

unified models hardware 128gb params general

Related Articles