What are the major open-weight model families in 2026, and who makes them?

beginner

← All Open-Source & Local LLMs questions

Re-explain

Around ten families from US, Chinese and French labs cover almost everything teams self-host. A family is one lab's ongoing series of models, not a single release.

From US labs:

Llama by Meta, custom community license
gpt-oss by OpenAI, Apache 2.0
Gemma by Google, Apache 2.0 since Gemma 4, custom Gemma terms before it
Phi by Microsoft, MIT
Olmo by Ai2, Apache 2.0, with open training data too

From Chinese labs:

Qwen by Alibaba, Apache 2.0
DeepSeek by DeepSeek, MIT
GLM by Z.ai, MIT. Z.ai was formerly Zhipu.
Kimi by Moonshot AI, modified MIT

France's Mistral AI makes Mistral, Mixtral and Magistral, under Apache 2.0 or research-only terms. IBM Granite and NVIDIA Nemotron are also common. So are TII's Falcon and MiniMax.

Apache 2.0 and MIT are permissive, while custom terms like Llama's add use limits. Chinese labs now ship many of the strongest permissive weights, often as large mixture-of-experts models that run only part of the model per token.

Each family ships sizes from roughly 1B to over 500B parameters, plus instruct and reasoning variants. Pick on license and size first, then test.

Rewriting in plainer words…

This answer doesn't lend itself to a diagram - it reads best . No credits were charged.

Why there's no diagram: “”

The interactive diagram is below the answer - jump to diagram ↓ · Below it, the related concept . Jump to it ↓

The diagram below the answer is the concept . Jump to it ↓

Tailored explanation · switch back to · ·

What should the new diagram focus on?

Guided practice for Open-Source & Local LLMs One question at a time. Answer out loud, get graded, see what you missed.

How well did you know this?

AI:

Saved in this browser - sign in to keep your review list.

How should your speech become text?

Listening… your words appear above as you speak - tap Stop when you're done.

Recording · cr - tap Stop & transcribe when you're done.

Transcribing with AI…

Voice:

Keep going - a few more words and AI can grade it.

Interview lens How interviewers actually use this question

Likely follow-ups

How do Apache 2.0, MIT and the Llama license differ if you ship a commercial product? Permissive licenses let you use, change and ship with notice. Custom terms add use policies and conditions, so read each model's actual license.
Many top open models are mixture-of-experts. What does that change when you host one? All experts must fit in memory, but only some run per token. Memory tracks total parameters and per-token compute tracks active ones.
A family spans 1B to over 500B parameters. How do you pick a size for your task and hardware? Start from what fits in memory after quantization. Try the smallest model that passes your eval, and move up only when it fails.
Would you worry about running a Chinese open-weight model inside a regulated company? Separate data flow from the rest. Local weights send no data home. The real checks are license, trained-in bias, company policy and file provenance.
Once you have a shortlist, how do you test which family is good enough for your use case? Build an eval from your own tasks, run candidates at the quantization you will ship, and weigh quality against latency and cost, not leaderboards.

What you can say

About ten open-weight families cover almost everything teams self-host. A family is one lab's ongoing series of models, not a single release.
From US labs you have Meta's Llama, OpenAI's gpt-oss, Google's Gemma, Microsoft's Phi, and Ai2's Olmo, which even publishes its training data.
From China there's Alibaba's Qwen, DeepSeek, Z.ai's GLM and Moonshot's Kimi, and from France, Mistral AI. IBM Granite and NVIDIA Nemotron show up a lot too.
Licenses matter as much as names. Most are Apache 2.0 or MIT, which are permissive, but Llama's custom license adds use limits.
Chinese labs now ship many of the strongest permissive weights, often as big mixture-of-experts models that run only part of the model per token.
Each family spans roughly 1B to over 500B parameters, so I'd shortlist on license and size first, then test on my own task before committing.

Weak answers to avoid

Lists model names without labs or licenses The question asks who makes them. Pair each family with its lab, and add the license, since that decides whether a company can ship it.
Calls every open-weight model free to use Apache 2.0 and MIT are permissive, but Llama uses custom terms with use limits. Treating them as the same tells the interviewer you'd miss a legal risk.
Leaves out the Chinese labs Qwen, DeepSeek, GLM and Kimi ship many of the strongest permissive weights. A US-only list sounds a few years out of date.
Treats a family as one model Llama or Qwen is a series with sizes from about 1B to over 500B plus instruct and reasoning variants. Name the family, then say how you'd pick the size.

Interview lens

Likely follow-ups, what you can say, and the weak answers to avoid.

Want a quick review of the fundamentals? See the Open-Source & Local LLMs cheatsheet.

← Back to all Open-Source & Local LLMs questions

Pro · $10/mo

48 of 55 Open-Source & Local LLMs answers are in Pro.

Full answers, code samples, and AI explanations that go simpler or deeper. Cancel anytime.

Full answers + code
AI explanations, simpler or deeper
1,000 AI credits / month
Cancel anytime

Get Pro → See plans & pricing →