Friendli

FriendliAI announced FriendliLink on September 21, 2026, a CLI tool that connects open-weight models to coding agents with a single `frlink on` command. The tool handles inference endpoint setup, authentication, model ID mapping, and reasoning configuration across supported agents including Claude Code, Codex, Cursor, FriendliLink includes an `on`/`off`/`status` workflow, lets developers browse available models and pricing directly from the CLI, and supports switching models without reconfiguring each agent. A saved configuration state allows developers to restore their previous setup when switching back. The tool runs on Friendli M

On April 15, 2026, FriendliAI announced support for the Anthropic Messages API across both Model APIs and Dedicated Endpoints. Developers can now run open-weight models — including GLM-5.1, MiniMax-M2.5, Kimi-K2.5, and DeepSeek-V3.2 — through the Messages API protocol (POST /v1/messages) by pointing existing Claude-bas The Messages API support includes multi-turn conversations, tool use, streaming, and extended thinking, distinguishing it from the OpenAI Chat Completions format. FriendliAI frames the addition as enabling significant cost reductions versus Claude, particularly for coding agents that drive large API costs. Applications

FriendliAI has entered a partnership with Aolani, a Singapore-founded neocloud, under which Aolani will supply GPU cloud infrastructure to support FriendliAI's inference platform, according to a September 10, 2026 press release distributed via ZAWYA/Media OutReach Newswire. The agreement is positioned to meet growing i The release frames the deal as a response to compute scarcity as more companies move AI workloads from experimentation to production deployment. It cites FriendliAI's origin in continuous batching research and its existing enterprise customers including LG, Kilo Code, and Liner, and includes a quote from Aolani CEO Nic

March 13, 2026 — FriendliAI has launched Friendli InferenceSense, an inference monetization platform purpose-built for GPU cloud operators.

Friendli InferenceSense - the "AdSense for GPUs" - Monetize Idle GPU Capacity with Inference "Assembly Line" for AI Factories

FriendliAI's official changelog documents an active Model APIs lifecycle through July 2026, with developers seeing frequent model additions and deprecations on the serverless platform. Recent dated entries include the deprecation of the Tool Assisted API on July 6, 2026, the removal of zai-org/GLM-5 on July 3, 2026, an Earlier in 2026 the changelog also surfaces removals such as deepseek-ai/DeepSeek-V3.1 (May 22), Qwen/Qwen3-30B-A3B (April 16), zai-org/GLM-4.7 (April 15), and MiniMaxAI/MiniMax-M2.1 (April 9), showing a steady rotation of open-weight model coverage on FriendliAI's Model APIs. The changelog UI additionally exposes filt