# Long-context reasoning

AWQ-quantized version of DeepSeek R1 0528, supports full-context-length operation on 8x80GB GPUs using vLLM.

QwenLong-L1 is a long-context reasoning model trained with reinforcement learning, demonstrating excellent performance across seven long-context document QA benchmarks.

A 4B-parameter large language model released by NVIDIA, supporting 128k tokens context length, optimized for reasoning, dialogue, and RAG tasks

AM Thinking v1 is a large language model developed by the A-M team based on Qwen 2.5-32B-Base, with enhanced reasoning capabilities and support for a context length of 132k tokens.

High-performance GGUF conversion version based on Qwen3-14B large language model, optimized for creative reasoning, deep dream logic, agent interaction, and multilingual instructions

This is a NEO Imatrix quantized version based on the Qwen3-4B model, using BF16 format MAX output tensors to enhance reasoning and output generation capabilities, supporting 32k context length.

Enhanced reasoning model based on Qwen-2.5 14B architecture, optimized for general reasoning and Q&A scenarios, supporting 128K context and 8K output

A merged model based on Llama 3 70B architecture, optimized for story generation and role-play, supporting long context windows

Granite-3.2-2B-Instruct is a 2-billion-parameter long-context AI model specifically fine-tuned for reasoning capabilities. Built upon Granite-3.1-2B-Instruct, it was trained using a mix of permissively licensed open-source datasets and internally generated synthetic data to enhance performance on reasoning tasks.

Theta-Lyrae-Qwen-14B is a 14-billion-parameter model based on the Qwen 2.5 14B modal architecture, optimized for general reasoning and Q&A capabilities, excelling in context understanding, logical reasoning, and multi-step problem-solving.

Galactic-Qwen-14B-Exp2 is a large language model based on the Qwen 2.5 14B architecture, focusing on enhanced reasoning capabilities, excelling in context understanding, logical reasoning, and multi-step problem solving.

RombUltima-32B is a fusion model that combines the strengths of Rombos-LLM-V2.5-Qwen-32b and Ultima-32B, optimizing reasoning capabilities, multilingual understanding, and multi-turn dialogue performance.

AWQ quantized version of DeepSeek R1 model, optimized for float16 overflow issues and supports efficient inference deployment

ModernBERT is a model fine-tuned on multi-task natural language inference (NLI) tasks, excelling in zero-shot classification and long-context reasoning.

Phi-3-Small-128K-Instruct is a 7-billion-parameter lightweight open-source model focused on high quality and strong reasoning capabilities, supporting 128K long context, and excelling in tasks such as commonsense reasoning, language understanding, mathematics, and coding.

Phi-3-Medium-128K-Instruct is a lightweight open-source model with 14 billion parameters, focusing on high quality and strong reasoning capabilities, supporting a 128K context length.

C4AI Command R+ is a 104B parameter multilingual large language model supporting Retrieval-Augmented Generation (RAG) and tool calling, optimized for tasks like reasoning, summarization, and Q&A.

Einstein-v4-7B is a large language model fully fine-tuned on diverse scientific datasets based on Mistral-7B-v0.1, specializing in STEM field tasks

A 32k-context fine-tuned version based on TinyLlama-1.1B, achieving long-context processing capability by increasing rope theta