Aethir Mesh正式開放：自有算力直供推理，11款頂級開源模型統一API接入

閱讀本文需 6 分鐘

Aethir Mesh已上線來自五大開源提供商的共計 11 款模型。

2026 年 6 月 3 日，Aethir 正式向所有開發者和 AI 智能體運營者開放 Aethir Mesh——這是 Aethir 自主研發的開源大模型 API 平台，所有推理任務直接運行於 Aethir 去中心化 GPU 基礎設施之上。

算力直供，而非路由中間層

理解 Aethir Mesh 的定位，需要先理解它與現有方案的本質差異。

以 OpenRouter 為代表的主流開源模型 API 平台，本質上是一個聚合路由層——將用戶請求轉發至 OpenAI、Anthropic、Google 等第三方推理提供商的伺服器，平台從中收取轉售溢價。用戶數據在這一過程中必然流經外部伺服器。

Aethir Mesh 的架構截然不同：開源模型直接運行在 Aethir 自有的去中心化 GPU 基礎設施之上——43 萬餘個 GPU 容器、覆蓋 94 個國家、200 餘個節點。Aethir 是實際執行計算的那一層，而非他人 API 的帳單封裝。這意味著無轉售溢價、無第三方數據路由、GPU 利用率維持在 95% 以上的企業級服務水準。

11 款模型，單一端點

Aethir Mesh 當前已上線來自五大開源提供商的共計 11 款模型，核心陣容包括：

• DeepSeek V4：1.6 萬億參數 MoE 架構，每步推理僅啟動其中一小部分，以接近前沿閉源模型的性能表現實現大幅低於市場價的推理成本，適合大多數智能體通用任務；

• Kimi K2.6：Moonshot AI 出品，1 萬億參數 MoE、每步 320 億活躍參數，原生支持大規模子智能體協同編排，專為多智能體流水線設計；

• GLM-5.1：Z.AI 出品，MIT 開源許可證，MoE 架構、每步 400 億活躍參數，超長上下文窗口專為長程任務設計，支持單次連續運行長達 8 小時、數千次工具調用；

• MiniMax-M2.5：2026 年 2 月發布，面向編程、智能體工具使用、網頁搜索及複雜辦公任務場景，推理速度達 206.7 tokens/秒；

• Qwen3.6-27B: Produced by Alibaba, with 270 billion parameters, Apache 2.0 license, native 262K Token context (scalable to millions), comprehensively outperforming the previous generation flagship Qwen3.5-397B on major coding benchmarks.

All models are accessed through a unified API endpoint, with developers managing a single key, single bill, single dashboard, eliminating the need to maintain multiple sets of authentication processes across platforms.

Integrated with Aethir Claw

For users of the Aethir Claw smart agent platform, Aethir Mesh is the final piece of their full-stack closed loop. After users write the Mesh API Key into the smart agent configuration on the Aethir Claw dashboard, all model inference requests from this smart agent will automatically be routed through Aethir Mesh and return results within the platform—VPS instances, skill execution layer, and model inference layer all share the same infrastructure, unified billing, unified monitoring, with the inference data never leaving the Aethir ecosystem.

For independent developers and enterprises not using Aethir Claw, Aethir Mesh can also serve as a standalone inference endpoint, suitable for Agentic AI application development, DePIN projects, and enterprise AI workflows.

Continuously Expanding Model Catalog

Aethir Mesh's model catalog is designed to dynamically expand with the open-source ecosystem. As high-performance open-source models are successively released and reach production stability, Mesh will continue to incorporate new models, allowing users to access the latest capabilities without switching API providers or renegotiating access terms.

Technical Documentation: docs.aethir.com/aethir-mesh

Get Started Now: https://agent.aethir.com/twitterCNmesh