Local AI Agents, Meta Muse Wonder + a GPT-4 API Deadline Worth Knowing

AI and SaaS Roundup · 2026-10-14

Guys, two new models worth knowing about if you’re building agents or coding tools. Plus, OpenAI has just put an end date on some familiar API models. 👀

An agent that runs on your own machine? Liquid AI has a new option

Liquid AI released LFM2.5-24B-A2B on 13 October. The weights are downloadable, and it supports tool calling for agents running on your own hardware. There’s support for llama.cpp, MLX, vLLM and SGLang, including Apple Silicon through MLX.

The team’s demo runs Hermes Agent locally to put together a briefing and handle follow-up questions. Think a private document assistant or an agent working through local project files as possible things to build with it.

It’s still a 24B model. The “2B active” description doesn’t mean it’s a tiny 2B download, so hardware and quantisation matter. It’s open-weight under Liquid’s own licence, rather than plain Apache 2.0. The free commercial-use licence has a $10 million annual-revenue threshold.

Source: Liquid AI: release announcement, downloads and linked model licence

Meta has new coding models to try: Muse Wonder

Meta launched Muse Wonder and Muse Wonder Full on 12 October. Muse Wonder is available through the Meta Model API and Muse Code. Full is available to Muse Code subscribers.

The focus is longer jobs: keeping useful context between conversations and working through changes, tests and retries. Meta calls those capabilities Active Memory and Active Focus. Its research post includes a demo of the model building a playable 3D kitchen game from images.

Another option for builders comparing coding and agent tools. The launch benchmarks are Meta’s own evaluations, rather than a guarantee of how it will perform on your project.

Source: Meta: Muse Wonder release and developer availability

Still using GPT-4o or GPT-4.1 in your app? There’s a shutdown date

OpenAI announced today, 14 October, that several GPT-4 API models will be removed on 14 January 2027. The notice covers GPT-4o snapshots from August and November 2024, GPT-4o mini from July 2024, and GPT-4.1 and GPT-4.1 mini from April 2025. Fine-tuned versions follow the same schedule.

The recommended replacement in the notice is gpt-6-luna. This is about the API, so it affects apps calling those models; it isn’t an announcement that your ChatGPT subscription is ending.

The exact model IDs and dates are in the source below. Useful if an older model is still quietly powering part of your SaaS.

Source: OpenAI: 14 October GPT-4 API deprecation notice