Model Hub

Qwen3 30B-A3B Thinking — Alibaba Qwen Mixture-of-Experts LLM (30B total, 3B activated per token) tuned for deep step-by-step reasoning with explicit thinking traces, excelling at math, code and logic. Apache 2.0.

SwinIR Classical x4 - Swin Transformer image-restoration model for 4x super-resolution, reconstructing sharp high-resolution detail from low-resolution images. Works at any input size. Apache 2.0.

Image-to-video diffusion, 14B parameters, 480P clips.

Kokoro 82M — hexgrad lightweight 82M-parameter text-to-speech model (built on StyleTTS 2) delivering natural, high-quality voices at very low compute cost, ideal for fast and on-device synthesis. Apache 2.0.

Qwen3-Coder 30B-A3B Instruct — Qwen's MoE coding model (48 layers, 128 experts top-8, 3B active) tuned for agentic coding, with native 262k-token context. Apache 2.0.

GLM-4.1V-9B-Thinking — vision-language reasoning model (image+text → text with <think> traces). Dynamic-resolution vision, first VLM on NeuroBrix; validated 4/4 execution modes token-identical.

Open text-to-video generation, MMDiT architecture, 11B parameters.

Swin2SR Classical ×2 — Swin Transformer V2 image super-resolution model that upscales images 2× while restoring fine detail; the classical-SR variant for clean low-resolution inputs. Apache 2.0.

Granite Speech 3.3 8B — IBM speech-language model that pairs a speech encoder with the Granite 3.3 LLM for automatic speech recognition and speech-to-text translation across English, French, German, Spanish and Portuguese. Apache 2.0.

Real-ESRGAN ×4 — xinntao practical image and photo super-resolution model (GAN-based) that upscales 4× while removing real-world noise, blur and compression artifacts. BSD-3 license.

Janus-Pro-7B unified multimodal (image generation). Validated engine surface: PyTorch branch (compiled + sequential, byte-identical); image-autoregressive on the Triton engine is pending and refused with a clear message until then.

Text-to-video generation at native 720x1280, 88-frame clips, 3D-RoPE diffusion transformer with T5 conditioning.