# Long context processing

Voxtral Mini is an enhanced version based on Ministral 3B, with advanced audio input capabilities and excellent performance in speech transcription, translation, and audio understanding.

Lucy is a 1.7-billion parameter model focused on intelligent web search and lightweight browsing. It is built on Qwen3-1.7B and optimized to run efficiently on mobile devices.

A language model merged from multiple Qwen 3 fine-tuned models, integrating code and mathematical abilities to provide powerful language processing capabilities.

EXAONE 4.0 is a multilingual large language model that integrates non-inference mode and inference mode, supporting English, Korean, and Spanish, and has the ability to invoke agent tools.

The EXAONE 4.0 model integrates non-inference mode and inference mode, combining the excellent usability of EXAONE 3.5 with the advanced inference capabilities of EXAONE Deep. It supports three languages: English, Korean, and Spanish, providing stronger support for intelligent applications.

EXAONE 4.0 is a multi-functional large language model that integrates non-reasoning and reasoning modes, supports English, Korean, and Spanish, and has the ability to use agent tools.

A.X 3.1 Light is a lightweight large language model (LLM) optimized for Korean understanding and enterprise deployment, fully independently developed by SKT.

SmolLM3 is a language model with 3 billion parameters, aiming to break the performance boundary of small models. It supports 6 languages, has advanced reasoning ability and long context processing ability, and is a fully open-source model.

A.X 4.0 Light is a lightweight large language model developed by SKT AI Model Lab, built on Qwen2.5 and optimized for Korean understanding and enterprise deployment.

The LongWriter-Zero-32B quantized model is based on the THU-KEG/LongWriter-Zero-32B base model, supports both Chinese and English, and is suitable for long context scenarios such as reinforcement learning and writing.

The LongWriter-Zero-32B Quantized Model is a multilingual model that undergoes static quantization based on the original model. It is suitable for long context scenarios such as reinforcement learning and writing.

SmolLM3 is an open-source language model with 3 billion parameters, supporting 6 languages and possessing powerful inference and long context processing capabilities.

Jan Nano is a small language model fine - tuned based on the Qwen3 architecture, designed for local and embedded environments, combining efficiency and long - context processing capabilities.

The GPTQ quantized version of the DeepSeek-R1-0528 model, using a quantization scheme of Int4 + selective Int8, which reduces the file size while ensuring the generation quality.

Zlatorog-12B-Instruct-Beta is a large language model fine-tuned based on Mistral Nemo 12B, designed specifically for Slovenian language-related research, supporting instruction following, reasoning, and tool-enhanced generation tasks.

Phi-4-reasoning is an advanced reasoning model developed by Microsoft, fine-tuned based on Phi-4, focusing on improving reasoning abilities in fields such as mathematics, science, and coding.

Qwen3 is a series of advanced language models developed by the Qwen Team, excelling in advanced reasoning, dialogue, instruction following, and agent applications.

Bamba-9B-v2 is a decoder-only language model built on the Mamba-2 architecture, focusing on text generation tasks and outperforming Llama 3.1 8B in performance.

A Greek instruction-tuned large language model based on Llama-3.1-8B, enhancing Greek language capabilities and supporting multilingual tasks

Llama-Krikri-8B-Base is a large Greek foundational language model built upon Llama-3.1-8B, extending Greek language capabilities through continued pretraining while maintaining English proficiency.

A multi-purpose large language model that combines the advantages of reasoning models and long context models

DeepSeek-R1 is the first-generation inference model, which performs excellently in mathematics, code, and reasoning tasks, and its performance is comparable to that of OpenAI-o1.

Lucie-7B is a multilingual causal language model with 7 billion parameters jointly built by LINAGORA and OpenLLM-France. It is based on the Llama-3 architecture and pre-trained on 3 trillion tokens of multilingual data.

Based on the ablated version of the large language model, with better performance than the original model, supporting Chinese, English, and long context processing.

AI21 Jamba 1.5 is a series of advanced foundation models with powerful long context processing capabilities and efficient inference speed, suitable for various business scenarios.

The Jais series are specialized bilingual large language models for Arabic processing, adapted for Arabic through pre-training based on the Llama-2 architecture

The Jais series is a bilingual large language model specifically optimized for Arabic while maintaining strong English capabilities. This 590M parameter version is the smallest model in the series, suitable for lightweight applications.

The Jais series is a set of English-Arabic bilingual large language models, optimized for Arabic while maintaining strong English capabilities, with models ranging from 590 million to 70 billion parameters.

The Jais series is a large English-Arabic bilingual language model optimized specifically for Arabic. It undergoes adaptive pre-training based on the Llama-2 architecture and has powerful bilingual processing capabilities.

Meltemi 7B Instruct v1.5 is a Greek instruction fine-tuned large language model improved based on Mistral 7B, focusing on Greek natural language processing tasks.

Meltemi is a large-scale Greek basic language model trained by the Athens Center for Research and Innovation, built on Mistral 7B and focused on Greek language processing capabilities.

Aya 23 is an instruction-fine-tuned open-weight research version model with highly advanced multilingual capabilities, supporting 23 languages.

Evo is a biological foundation model capable of long context modeling and design. It uses the StripedHyena architecture and can model sequences at single nucleotide and byte-level resolution.

The Yi series of models are next-generation open-source large language models trained from scratch by 01.AI. They support bilingual (Chinese and English) and perform excellently in language understanding, common-sense reasoning, reading comprehension, etc.

An extended mind Transformer model adjusted based on the Mosaic ML architecture, supporting external memory library retrieval and attention functions