HPC-AI

HPC-AI launched its Model APIs product, providing developers access to open models through simple APIs. The launch post states the initial lineup included Kimi K2.5 and another open model, with the goal of enabling AI-assisted coding tools, personal AI agents, and AI-powered applications to be built within minutes. To celebrate the launch, HPC-AI offered $2 in free credits to all users. The post estimates this could support approximately 4.3M input tokens and 1.07M output tokens under a typical agent workflow with 50% cache hit rate. The company also announced a rebate program for higher-volume API consumers.

HPC-AI announced the integration of Zhipu AI's GLM-5.1 into its Model APIs platform. According to the blog post, GLM-5.1 is positioned as a flagship agentic engineering model capable of maintaining productivity for up to 8 hours on complex multi-step tasks, with claimed state-of-the-art results on SWE-bench Pro and Terminal-Bench 2.0 benchmarks. The model uses a 744B parameter Mixture-of-Experts architecture. The blog highlights a demo in which GLM-5.1 reportedly constructed a complete Linux desktop environment from scratch in 8 hours. HPC-AI is offering GLM-5.1 through its Model APIs dashboard with what it describes as a competitive pricing structure optimized for high-volume agentic workflows and context caching, alongside enterprise-grade uptime and low-latency inference infrastructure.

HPC-AI Tech's official blog shows a concentrated push of developer-tooling integrations against its unified Model APIs throughout June and July 2026, including tutorials for Cursor, Claude Code, Codex, Aider, AutoGen, AutoGPT, Cline, Crew AI, Dify, Droid, Flowise, Goose, Grok CLI, Hammer AI, Helicone, and Kilo Code dat A June 28, 2026 post also confirms HPC-AI Tech will exhibit at ICML 2026 in Seoul (July 6–11, Booth B301), promoting its AI infrastructure platform alongside an ICML Author Incentive Program, which indicates ongoing investment in researcher-facing outreach and developer relations around its inference platform. No new m