catalog: add 27 current models and sort the catalog

[shacortes]

The catalog had four models, the newest from mid-2025. This adds twenty-seven that people are actually running locally today (September 2026), so there are boards to post results against.

Every entry was checked against the Hugging Face API before it was written: the repo exists, the sourceUrl resolves to the official org's card, and params / architecture / activeParams come from the model's own config.json and card, not from the name.

Model	Params	Arch
Qwen3.8-27B	27B	dense
Qwen3.8-Flash-Next	125B	MoE, 6B active
Qwen3.8-2.4T-A95B	2.4T	MoE, 95B active
Qwen3.6-27B	27B	dense
Qwen3.6-35B-A3B	35B	MoE, 3B active
Qwen3.5-0.8B	0.8B	dense
Qwen3.5-2B	2B	dense
Qwen3.5-4B	4B	dense
Qwen3.5-9B	9B	dense
Qwen3.5-27B	27B	dense
Qwen3.5-35B-A3B	35B	MoE, 3B active
Qwen3.5-122B-A10B	122B	MoE, 10B active
Qwen3.5-397B-A17B	397B	MoE, 17B active
Qwen3-4B-Instruct-2507	4B	dense
Qwen3-Next-80B-A3B-Instruct	80B	MoE, 3B active
Qwen3-Coder-Next	80B	MoE, 3B active
Gemma 4 12B IT	12B	dense
Gemma 4 26B A4B IT	26B	MoE, 4B active
gpt-oss-20b	21B	MoE, 3.6B active
Mistral Small 3.2 24B Instruct	24B	dense
Granite 4.2 8B	8B	dense
Granite 4.2 30B	30B	dense
Nemotron 3.5 Lightning 30B A3B	30B	MoE, 3B active
LFM2.5-2.6B	2.6B	dense
LFM2.5-8B-A1B	8B	MoE, 1B active
MiniCPM5-2B	2.5B	dense
GLM-5.3-Flash	320B	MoE, 18B active

Notes on the choices:

• Quants are per-model, not a blanket list. Only quantizations that actually ship somewhere are listed — GGUF where an official or widely used GGUF repo exists, int4/int8 where an OpenVINO or AWQ/GPTQ build exists (OpenVINO/Qwen3.8-27B-int4-ov, OpenVINO/LFM2.5-8B-A1B-int4-ov), mxfp4 for gpt-oss-20b because that is its native format.
• Small models are in on purpose. MiniCPM5-2B, LFM2.5-2.6B, and the Qwen3.5 sub-10B sizes are the ones that fit an NPU or an iGPU, which is where the Core Ultra boards have nothing to compare today.
• GLM-5.3-Flash is the one entry that will not fit a card — it is here for the Xeon and multi-card rigs, with only int4 and q4_k_m listed.
• Logos: reused qwen.svg and gemma.svg for the new Qwen and Gemma entries. The other families have no SVG in frontend/public/logos/models/, so they fall back to a monogram tile in brandColor, in the same muted register as the existing four.
• The MODELS array is kept alphabetical by name, so new entries slot in by name order rather than always appending to the end.

Checks run locally:

npm --prefix frontend run catalog:validate   # ✅ 46 hardware, 31 models, 14 quants, 7 runtimes
npm --prefix frontend run typecheck          # clean

No seed migration in this PR — per frontend/src/catalog/README.md that is a maintainer step.

[vercel]

@shacortes is attempting to deploy a commit to the Community Labs Team on Vercel.

A member of the Team first needs to authorize it.

[github-actions]

No result files to import.

Site / sign up · Workflow details and retry

[shacortes]

Why doesn't Claude listen 😔

[jackwsmth]

Hey, think this looks okay so far.

I can use the models you've added to the catalog here to the backend as well.

[shacortes]

Hey, think this looks okay so far.

I can use the models you've added to the catalog here to the backend as well.

I was thinking of adding a few more, as it's missing the qwen3.6 ones. And some other models.

[jackwsmth]

If you want to keep working on it, feel free to re-open the PR and ping me when you think it's in a suitable spot 🙂

[shacortes]

If you want to keep working on it, feel free to re-open the PR and ping me when you think it's in a suitable spot 🙂

There's a slight problem. If I add too many, searching for a specific one is like searching for one's country on a drop-down list, not hard but much more tedious than being able to start searching. I'd open this PR, drop the models and merge, and after that, if it could be done, implement something like a search bar of sorts to look for a specific model. Or something similar to the hardware selector, whatever's most preferred.
Also if possible, could also make the list alphabetically sorted.

Reopening and working on it.

[jackwsmth]

Appreciate the notes. I'll take a look at the model list you've added here, and if all looks good I can merge it and add to the DB.

I'll consult with @horacechoi on the UI, and we can determine search/dropdown specifics etc. Might be easier to sort by model family.

[shacortes]

I think I'm done. I did move models around so feel free to revert if needed. This allows it to more or less have the models sorted alphabetically, though I do think it'd look much better if programatically done. Well, technically it does look nice if it's sorted on the index.ts, but who knows.
@jackwsmth
(I don't like pinging)

[shacortes]

How would the quants be listed? Or do I keep the "simple" ones and that's it? Wondering because there's more quant types other than q4 and variants, even.

[jackwsmth]

How would the quants be listed? Or do I keep the "simple" ones and that's it? Wondering because there's more quant types other than q4 and variants, even.

I think the quants you've added are perfect for now. If we get requests to add more, we can do so.

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

Project	Deployment	Actions	Updated
intelinside	Ready	Preview	Sep 16, 2026 5:32pm UTC

[jackwsmth]

Appreciate you contributing @shacortes! 🎉

Please feel free to open any other issues/PRs if you think there is something we can improve on 😄