Author: Aethir
On June 3, 2026, Aethir officially opened Aethir Mesh to all developers and AI agent operators—Aethir’s proprietary open-source large model API platform, where all inference tasks run directly on Aethir’s decentralized GPU infrastructure.
Direct hash power supply, not via routing intermediaries
To understand the positioning of Aethir Mesh, one must first grasp its fundamental differences from existing solutions.
Mainstream open-source model API platforms like OpenRouter essentially function as aggregation routing layers—forwarding user requests to third-party inference providers such as OpenAI, Anthropic, and Google, while charging a markup on the resale. User data inevitably passes through external servers in this process.
The Aethir Mesh architecture is fundamentally different: the open-source model runs directly on Aethir’s own decentralized GPU infrastructure—over 430,000 GPU containers spanning 94 countries and more than 200 nodes. Aethir is the actual layer executing computations, not a billing wrapper for third-party APIs. This means no reseller markups, no third-party data routing, and enterprise-grade service with GPU utilization consistently above 95%.
11 models, single endpoint
Aethir Mesh currently features 11 models from five open-source providers, including the core lineup:
• DeepSeek V4: A 1.6 trillion parameter MoE architecture that activates only a small portion of parameters during each inference step, achieving performance close to leading proprietary models at significantly lower costs than market rates, ideal for most general-purpose agent tasks;
• Kimi K2.6: Developed by Moonshot AI, a 1-trillion-parameter MoE model with 32 billion active parameters per step, natively supporting large-scale sub-agent collaborative orchestration, designed specifically for multi-agent pipelines;
• GLM-5.1: Developed by Z.AI, licensed under the MIT open-source license, featuring a MoE architecture with 40 billion active parameters per step, an ultra-long context window designed for long-range tasks, and support for continuous operation up to 8 hours with thousands of tool calls;
• MiniMax-M2.5: Released in February 2026, designed for programming, agent tool usage, web search, and complex office tasks, with a reasoning speed of 206.7 tokens/second;
• Qwen3.6-27B: Developed by Alibaba, 27 billion parameters, Apache 2.0 license, native 262K token context (scalable to millions), outperforming the previous flagship Qwen3.5-397B across major coding benchmarks.
All models are accessible through a unified API endpoint, allowing developers to manage a single key, single billing account, and single dashboard without maintaining multiple authentication processes across platforms.
Full-stack integration with Aethir Claw
For Aethir Claw agent platform users, Aethir Mesh is the final piece of the full-stack闭环. After users enter their Mesh API Key into the agent configuration of the Aethir Claw dashboard, all model inference requests from that agent are automatically routed through Aethir Mesh and return results within the platform—VPS instances, skill execution layers, and model inference layers share the same infrastructure, unified billing, and unified monitoring, with all inference data remaining entirely within the Aethir ecosystem.
For independent developers and enterprises not using Aethir Claw, Aethir Mesh can also be directly integrated as an independent inference endpoint, suitable for Agentic AI application development, DePIN projects, and enterprise-grade AI workflows.
Continuously expanding model directory
The Aethir Mesh model catalog is designed to dynamically expand with the open-source ecosystem. As high-performance open-source models are released and reach production stability, Mesh will continuously integrate new models, enabling users to access the latest capabilities without switching API providers or renegotiating access terms.