Models

Qwen-Image-2512

You can use Qwen‑Image‑2512 as a high‑quality image generator for production apps, design tools, and creative workflows that need photorealistic people, complex scenes, and reliable typography in English and Chinese.…

Apertus-70B-2509

Apertus‑70B‑2509 is a 70B-parameter, fully open multilingual transformer from the Swiss AI Initiative, trained on 15T compliant tokens and supporting 1,800+ languages with competitive open‑weight benchmark performance.

GPT OSS 20b

gpt-oss-20b is a 21B-parameter open-weight MoE reasoning model from OpenAI with ~4B active parameters, a 128k context window, and native support for chain-of-thought, tools, and structured outputs under Apache 2.0.

Llama 3.1 8B Instruct

Llama-3.1-8B-Instruct is an 8B-parameter multilingual chat and instruction-following model from Meta with a 128k context window, strong tool usage, and efficient performance for real-time assistants.

Mistral Small 4 119b

Mistral-Small-4-119B-2603 is a 119B-parameter multimodal MoE model with only 6.5B active parameters per token, delivering top-tier reasoning, coding, and vision performance with a 256k-token context window.

Mistral-Small-3.2-24B-Instruct-2506

Mistral-Small-3.2-24B-Instruct-2506 is a 24B-parameter instruction-tuned model from Mistral that improves instruction following, reduces repetition, and offers robust function calling for production-grade assistants.

Llama 3.3 70B Instruct

Llama 3.3 70B Instruct is Meta’s multilingual, instruction-tuned 70B text model for chat, coding, reasoning, and tool-enabled assistants. It supports 128K context, eight officially supported languages, and commercial use under…

Qwen3.5-9B

Qwen3.5-9B is a 9B-parameter, open-weight multimodal foundation model from Alibaba Cloud that delivers strong reasoning, coding, and vision-language performance with a 262K-token native context window.

Qwen3-Coder-Next

Qwen3-Coder-Next is an open-weight coding model built for coding agents and local development, combining 80B total parameters with just 3B active parameters for efficient deployment. It supports 256K context and…

Qwen3-Embedding-8B

The model can be used with sentence-transformers or Hugging Face Transformers, with both integration paths documented on the official model card.

Qwen3-Reranker-4B

Qwen3-Reranker-4B is a 4B text reranking model built to improve retrieval precision in multilingual and code search workflows. It supports 32K context, 100+ languages, and instruction-aware ranking, making it well…

Qwen3-VL-32B-Instruct

supports seamless switching between a “thinking mode” for complex math/code/logic and a “non‑thinking mode” for efficient dialogue

GPT OSS 120B

gpt-oss-120b is OpenAI’s flagship open-weight Mixture-of-Experts language model with about 117B parameters and 5.1B active per token, optimized for high‑reasoning, agentic production workloads on a single 80GB GPU and released…

Qwen3.5-122b

Qwen3.5-122B-A10B is a powerful open-weight Mixture-of-Experts (MoE) model from Alibaba's Qwen team, featuring 122 billion total parameters with only 10 billion active per token for efficient performance.

Holo3‑35B‑A3B

Holo3‑35B‑A3B is a 35B (3B active) open‑weight multimodal MoE model from H Company, optimized for computer‑use agents that read screens, understand UIs, and plan reliable multi‑step actions with a 64k…

Chandra OCR 2

chandra‑ocr‑2 is a 4B‑parameter, layout‑aware OCR model from Datalab that converts complex documents into structured Markdown/HTML/JSON across 90+ languages, achieving SOTA olmOCR scores with 2× the throughput of Chandra 1.

Z Image Turbo

Z‑Image‑Turbo is a 6B‑parameter, ultra‑fast text‑to‑image model from Tongyi‑MAI that generates photorealistic, bilingual images in under a second using just eight diffusion steps, even on 16 GB consumer GPUs.

Qwen3.5 35B A3B

Qwen3.5‑35B‑A3B is a 35B (3B active) sparse‑MoE multimodal model from Alibaba that offers 262k–1M context, strong reasoning and vision performance, and Apache‑2.0 open weights tuned for efficient single‑GPU deployment.

GLM OCR

GLM‑OCR is a 0.9B‑parameter multimodal OCR model that uses a CogViT vision encoder and GLM decoder with Multi‑Token Prediction to deliver state‑of‑the‑art document parsing accuracy while remaining small enough for…

Nemotron Cascade 2 30B-A3B

Nemotron‑Cascade‑2‑30B‑A3B is a 30B (3B active) hybrid Mamba–Transformer MoE model from NVIDIA with 262k+ context and gold‑medal IMO/IOI 2025 performance, tuned for dense reasoning and agentic coding on a single…

NVIDIA Nemotron 3 Super 120B-A12B-NVFP4

NVIDIA Nemotron‑3‑Super‑120B‑A12B‑NVFP4 is a 120B (12B active) LatentMoE hybrid Mamba‑Transformer model with up to 1M‑token context, NVFP4 efficiency, and state‑of‑the‑art performance on agentic reasoning, coding, and long‑horizon planning tasks.