The Open Weights
LatestModelsLeaderboardsCompanies
Subscribe
The Open Weights

The daily record of open-source AI. New model releases, leaderboards, and what's coming next — written for people who ship.

Refreshed every 12 hours

Discover

  • Latest releases
  • New today
  • Trending models

Browse

  • All models
  • Companies
  • Categories
  • Leaderboards

About

  • About
  • Editorial policy
  • Hardware estimates
  • RSS feed
  • llms.txt
  • Newsletter

© 2026 The Open Weights. An independent publication.

PrivacyTermsSMSAggregated by Claude · curated by humans.
LatestOrnith Ai1.5
Ornith AiText / LLM

Ornith 1.5 Brings a Lean 35B MoE to Open Weights

The MIT-licensed model activates just 3B parameters per token, borrowing the Qwen3.5 MoE design for text and reasoning tasks.

Aug 18, 2026
UpdateMIT
Ornith-1.5-122B

Ornith AI has published Ornith-1.5-35B-A3B, a mixture-of-experts language model that leans on sparsity to keep inference costs down. While the model holds 35 billion parameters in total, only about 3 billion are active for any given token, according to its Hugging Face repository.

The release is built on the Qwen3.5 MoE architecture and targets both general text generation and reasoning. It ships under an MIT license, one of the more permissive options for commercial and research use, which lowers the barrier for developers who want to fine-tune or deploy it without restrictive terms.

Why it matters

The appeal of a 35B/3B-active setup is efficiency: MoE models can offer the knowledge capacity of a larger network while running at the speed and memory footprint closer to a much smaller dense model. That tradeoff has become a popular recipe for teams building on top of Qwen's architecture.

  • Total parameters: 35B, with roughly 3B active per token
  • Architecture: Qwen3.5 MoE
  • License: MIT
  • Focus: text generation and reasoning

As a smaller, incremental release, Ornith 1.5 is less about headline benchmarks and more about giving the open-weights community another permissively licensed MoE to experiment with. Details such as context length and evaluation results were not specified in the release record, so prospective users will want to check the repository directly before building against it.

Sources

  • ornith-ai/Ornith-1.5-35B-A3B

    Hugging Face

    Visit
OlderAnt Research releases 4DAnyone for 4D human videoAntResearch · Image → Video · 3 months agoNewerOrnith 1.5 arrives as a 397B MoE multimodal modelOrnith Ai · Text / LLM · 3 months ago

Get the model

Hugging Face

Specs

Parameters35B · MoE
Active params3B active
Context window262K tokens
Size71.9 GB
PrecisionBF16
ArchitectureQwen3_5MoeForConditionalGeneration
LicenseMIT
Downloads180K
Likes262

Can you run it?

Runs on a 24 GB GPU at 4-bit.

  • BF16 (as published)

    80 GB GPU (A100 / H100) · 128 GB Mac Studio

    73.6 GB
  • 8-bit

    48 GB GPU (RTX 6000) or 2×24 GB · 64 GB Mac

    39.8 GB
  • 4-bit

    24 GB GPU (RTX 3090 / 4090) · 32 GB Mac

    23.3 GB
Your machine
GGUF builds
  • Mixture-of-experts: all 36B parameters must be resident even though only 3B are active per token — active count affects speed, not memory.
  • KV cache sized for 8,192 tokens of context; longer prompts need proportionally more.

Based on Hugging Face weights metadata. Assumes an 8K context and ~1 GB runtime overhead; actual needs vary. How we estimate


Modalities

Text / LLMReasoning
7 versions — view changelog

The Weekly Weights

Every open release that mattered, one email a week.

0 comments

No comments yet. Be the first to weigh in.

More from Ornith Ai

All Ornith Ai releases →
Ornith-1.5-122B
Ornith Ai/Text / LLMWorkstation GPU

Ornith AI releases 122B MoE multimodal model under MIT

The Qwen3.5-based Ornith-1.5 pairs a mixture-of-experts design with vision support and an unusually permissive license.

Oct 16, 2026
Ornith-1.5-122B
Ornith Ai/Text / LLMMulti-GPU server

Ornith 1.5 arrives as a 397B MoE multimodal model

The MIT-licensed release spans a 397B mixture-of-experts flagship plus 9B and 35B-A3B variants for lighter deployments.

Aug 18, 2026

More in Text / LLM

All Text / LLM →
Aion-1.0-Instruct-2.3B
Microsoft/Text / LLMRuns anywhere

Microsoft's Aion 1.0 Instruct is a compact 2.3B chat model

A small, multilingual instruction-tuned model aimed at chat and summarization lands on Hugging Face.

Nov 18, 2026
SenseNova-Loop-9B
SenseTime/Vision-LanguageRuns on a laptop

SenseTime Releases Looped 9B Multimodal Model

The company's new SenseNova-Loop-9B pairs vision-language capabilities with a recurrent looped-transformer design.

Nov 17, 2026
Hunyuan Hy4-flash
Tencent/Vision-LanguageMulti-GPU server

Tencent releases Hunyuan Hy4-flash multimodal MoE

The lightweight, mixture-of-experts model adds vision and text instruction-following to Tencent's open Hunyuan lineup.

Nov 17, 2026