Together AI is a cloud platform that provides fast and affordable access to leading open-source AI models through an API, along with infrastructure for fine-tuning and training custom models. Founded in 2022 by a team of AI researchers from Stanford, the company operates a high-performance GPU cluster optimized for inference and training of open-source models. Together AI offers API access to a wide selection of popular open-source language models including LLaMA, Mistral, Mixtral, DeepSeek, Qwen, and many others, as well as image generation, code, and embedding models. The platform is known for its competitive pricing and fast inference speeds, achieved through custom inference engine optimizations and efficient GPU utilization. Together AI provides several key services. Its Inference API enables developers to run open-source models with OpenAI-compatible endpoints, making it straightforward to integrate into existing applications. The Fine-tuning API allows users to customize models on their own data with support for full fine-tuning, LoRA, and QLoRA methods, all managed through a simple API or web interface. Together also offers dedicated GPU clusters for organizations that need guaranteed capacity and custom deployments. The platform supports function calling, JSON mode, streaming, and chat completion formats that are compatible with the OpenAI API specification, simplifying migration for developers already using OpenAI. Together AI has contributed to several open-source projects and research efforts in efficient AI training and inference. Pricing follows a pay-per-token model that varies by model size and type, with rates generally lower than many competing inference providers. The platform is used by startups, enterprises, and researchers who prefer open-source models with the flexibility to fine-tune and customize.

AI GPU云

Together AI 运营针对 AI 推理和训练进行了优化的高性能 GPU 集群。它为需要保障资源的组织提供专用 GPU 容量，以及在用户之间高效共享 GPU 资源以实现成本有效的模型服务的无服务器推理。

AI模型托管

Together AI 在优化的基础设施上托管和提供数百个开源 AI 模型。开发者可以通过共享推理 API 部署模型以实现经济高效的提供，或为有保证的容量配置专用端点，由平台处理所有基础设施管理。

AI训练平台

Together AI 提供托管的微调和训练基础设施，用于自定义开源模型。用户可以通过简单的 API 使用全量微调、LoRA 或 QLoRA 方法来微调模型，由 Together 处理 GPU 配置、分布式训练和优化。

LLM API

Together AI 通过 OpenAI 兼容的端点提供对广泛的开源语言模型目录的 API 访问，具有竞争性的定价和快速的推理速度。开发者可以通过标准化的 API 访问 LLaMA、Mistral 和 DeepSeek 等模型，支持流式处理、函数调用和 JSON 模式。

开源 LLM

Together AI 专门托管和提供开源语言模型，为来自 Meta、Mistral、DeepSeek 和其他开源提供商的模型提供快速且经济实惠的 API 访问。其平台使运行、比较和集成开源 LLM 变得容易，无需管理 GPU 基础设施。