Together AI

Together AI optimized the deployment of MiniMax M3, achieving 81–125% throughput improvements through architectural and engineering innovations.

ThunderAgent通过优化KV缓存管理，实现单节点吞吐量提升2倍，集群加速2.4倍，解决代理推理中的缓存抖动问题。

Together AI平台通过端点/部署/配置三元组与容量感知路由实现模型推理配置，支持零停机变更和流量实验。

Together AI与Moonshot AI合作推出2.8T参数开源模型Kimi K3，支持长上下文和视觉任务，开发者可直接通过其平台部署和微调。

Together AI平台通过自适应扩展端点，结合LLM推理特有的指标（如TTFT、GPU利用率），实现更高效的资源管理，减少过量和不足配置的成本。

Kimi K3是首个2.8万亿参数开源模型，支持1M上下文和高效推理，适用于复杂编程与知识工作。

Kimi K3在DeepSWE基准测试中以1/3成本超越Claude Fable 5，多次尝试后性能反超，但后者可靠性更高。

不同可靠性级别（99%、99.9%、99.99%）对应不同的故障域和架构要求，如99%需处理节点级故障，99.9%需跨数据中心部署，99.99%需多区域冗余。

YC初创公司可通过专用GPU集群快速获取计算资源，无需长期合约。Together AI与YC合作解决AI公司算力瓶颈问题。

Together AI推出的新推理平台使用户能高效部署和管理开放权重模型，兼顾控制、成本与性能。

Kimi K3在软件工程任务中性能与Claude Fable 5相当，但价格仅为后者35%。

Inkling是Thinking Machines Lab推出的多模态模型，支持高效推理和跨任务能力，Together AI提供生产级部署服务。

Together AI 获得 ISO 27001:2022 认证，证明其信息安全管理符合国际标准，增强客户对其平台安全性的信任。

Building high-quality, low-latency, scalable voice agents is now an engineering challenge requiring real-time response (<500ms), complex instruction handling, and tool calling — supported by Together AI’s infrastructure.

Together AI optimized their speech-to-text stack, achieving faster transcription speeds by using profile-aware TensorRT, optimizing the decoder loop, and improving CPU paths. They serve the two lowest-latency models, with the fastest model transcribing 20 hours of speech in under 10 seconds.

Together Inference Engine delivers 31% more TPS than next fastest OSS engine on same hardware, maintains 2× better TTFT at saturation. Performance gains come from full-stack optimization.

Together AI and Pearl Research Labs have partnered to reduce AI inference costs through technologies like FlashAttention-4 and ATLAS.

Violin is an open-source video translation tool developed by Together AI, using multimodal models to achieve high-quality video content localization.

Together AI has launched a new tool called Voice Finder, which helps developers quickly find the right voice for their app from over 600 voices.

Together AI launches DeepSeek-V4 Pro model with high-performance inference and multiple computing options.

文章介绍了Together AI的多项技术进展，包括FlashAttention-4、ATLAS加速器和Batch Inference API更新，显著提升了大规模推理效率。

Y Combinator与Together AI合作建立GPU集群，为初创公司提供更便捷的计算资源，但技术细节和工程实践指导不足。

GLM-5.2 模型在 Hugging Face 推理服务中可通过多个平台免费使用，但仅限于接下来的 6 小时。