Frontier open weights

The giants: open models you (probably) can't run at home.

Some of the most important open-weight models are simply too large for consumer hardware. They still matter: they set the benchmark ceiling, their licenses shape the ecosystem, and their distilled siblings are often the best models you can run. Here is what each one would actually take — and the realistic way to use it anyway.

Looking for something that runs on your machine? Try the hardware picker or browse the curated directory.

Qwen3.8-2.4T-A95B

Alibaba (Qwen team) · China · Released August 2026

Custom Qwen licence — not the Apache 2.0 that covers the 27B

The largest model Alibaba has published, and the top trending release on the Hub at launch. Same generation as the Apache-licensed 27B, under a different licence.

Size: 2.4T MoE (95B active)
Context: 262k tokens, native multimodal
What running it yourself actually takes: Roughly 1.2 TB at 4-bit. Multi-node territory. Community GGUF conversions exist and are mostly of academic interest unless you have a cluster.
Realistic access: Alibaba Cloud's API and the usual inference providers. FP8 weights are published alongside the BF16 ones if you are renting datacenter GPUs.
Runnable sibling: Qwen3.8-27B — dense, Apache 2.0, multimodal, and in the picker.

Official site →

Kimi K3

Moonshot AI · China · Released June 2026 (weights public on the Hub)

Moonshot Open License

A frontier-scale open-weight MoE whose sheer weight footprint makes it infrastructure rather than homelab software.

Size: 2.8T MoE
Context: 1M tokens, native multimodal
What running it yourself actually takes: Even aggressive quantization requires well over a terabyte of memory. Practical self-hosting means a multi-node datacenter GPU cluster.
Realistic access: Moonshot API or a third-party inference provider.
Runnable sibling: Kimi K2.x family for smaller agentic workloads.

Official site →

GLM-5.2

Z.ai (Zhipu) · China · Released June 2026

MIT

A very large MIT-licensed open-weight model aimed at frontier coding and reasoning.

Size: 744B
Context: 1M tokens
What running it yourself actually takes: Hundreds of gigabytes even when quantized; multi-GPU workstation or datacenter territory.
Realistic access: Z.ai API or third-party inference providers.
Runnable sibling: Smaller GLM-family checkpoints remain practical local options.

Official site →

Llama 4 Maverick

Meta AI · United States · Released 2025

Llama 4 Community License

Meta's frontier-tier open-weight MoE: relatively light compute per token, but very heavy memory requirements.

Size: 400B MoE (~17B active)
Context: 1M tokens
What running it yourself actually takes: Roughly hundreds of gigabytes at practical quantizations; a single consumer GPU is not enough.
Realistic access: Hosted inference or Meta-supported API access.
Runnable sibling: Llama 4 Scout is smaller but still workstation-class.

Official site →

DeepSeek V4 Pro

DeepSeek AI · China · Released 2026

MIT

Frontier-scale DeepSeek model for code, math and agents under a permissive license.

Size: 1.6T MoE (49B active)
Context: 1M tokens
What running it yourself actually takes: The full checkpoint needs several hundred gigabytes to more than a terabyte depending on precision and runtime overhead.
Realistic access: DeepSeek API or third-party hosting.
Runnable sibling: DeepSeek V4 Flash — 304B-A13B, MIT, and in the picker for machines with 128 GB or more.

Official site →