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.
LatestNanbeige4.3-33B
NanbeigeText / LLM

Nanbeige 4.3 Arrives as a 33B Bilingual Model

A dense English–Chinese language model lands under Apache 2.0, aimed at teams that want permissive licensing at mid-size scale.

Oct 24, 2026
NotableApache 2.0
Nanbeige4.3-33B

Nanbeige has released Nanbeige4.3-33B, a dense bilingual language model that handles both English and Chinese. The model is published on Hugging Face under the permissive Apache 2.0 license, making it available for commercial and research use without the usage restrictions that accompany many open-weight releases.

At roughly 33 billion parameters, the model sits in a useful middle tier. It is a dense architecture rather than a mixture-of-experts design, which means every parameter is active on each forward pass. That tends to make deployment behavior more predictable, even if it forgoes the compute efficiencies that MoE routing can offer at larger scales.

Why it matters

The combination of a 33B dense model and an Apache 2.0 license is the notable part here. Many capable open models in this size range carry custom or community licenses with commercial caveats, so a cleanly permissive option gives developers more freedom to fine-tune, redistribute, and build products on top of the weights.

  • Dense 33B architecture, text-only
  • Bilingual coverage across English and Chinese
  • Apache 2.0 licensing with no added usage restrictions

Nanbeige has not published context length or benchmark details in this record, so teams evaluating the model will want to check the model card directly for configuration specifics and any accompanying evaluation results before committing to it.

Sources

  • Nanbeige/Nanbeige4.3-33B

    Hugging Face

    Visit
OlderIBM's Granite Speech 5.0 Adds Speaker AttributionIBM · Speech → Text · 4 days ago

Get the model

Hugging Face

Specs

Parameters33B
Context window524K tokens
Size66.2 GB
PrecisionBF16
ArchitectureLlamaForCausalLM
LicenseAPACHE-2.0
Downloads3.8K
Likes1.2K

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

    68.9 GB
  • 8-bit

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

    37.8 GB
  • 4-bit

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

    22.6 GB
Your machine
GGUF builds
  • 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 / LLM
3 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 Nanbeige

All Nanbeige releases →
Nanbeige4.3-33B
Nanbeige/Text / LLMRuns on a laptop

Nanbeige 4.2 arrives as a compact 3B bilingual model

The new English-Chinese language model targets efficient deployment in a small parameter footprint.

Jul 21, 2026
Nanbeige4.3-33B
Nanbeige/Text / LLMRuns anywhere

Nanbeige Releases 3B Chinese-Enhanced Language Model

The new Llama-based model was trained from scratch on 3.5 trillion tokens of Chinese and English data to enhance its bilingual capabilities.

Feb 10, 2026

More in Text / LLM

All Text / LLM →
Flint
RecursiveSI/Text / LLMConsumer GPU

Recursive releases Flint, an Apache-2.0 chat model

The new conversational LLM ships under a permissive license with community GGUF builds already available for local use.

Oct 21, 2026
Raon-1-mini (preview)
KRAFTON/Text / LLMConsumer GPU

KRAFTON releases Raon-1-mini preview under Apache 2.0

The Korean game studio behind PUBG puts out a compact, permissively licensed language model as an early preview.

Oct 21, 2026
DeepSeek-V4.1-Flash (1017)
DeepSeek/Vision-LanguageDatacenter

DeepSeek V4.1-Flash Adds Vision and Reasoning

The new MIT-licensed checkpoint folds image understanding and reasoning into DeepSeek's fast, mixture-of-experts Flash line.

Oct 17, 2026