🐟 New AI releases
Notable AI launches from the last ~30 days, with source links. Not yet tested by us.
-
Beam (open-weight model) — Reflection AI
Open-weight model for coding, reasoning and long agent tasks, Apache 2.0, free to start on Reflection's platform.
-
Muse Spark Open — Meta
Meta released weights for an 816B-parameter multimodal model built for coding and agentic work.
-
Nemotron 3.5 Super VL & Bolt — NVIDIA
Open models for always-on agents, plus small 'Bolt' models for local PCs.
-
Deep Review (public release) — Google Research
Agentic system that checks claims and reasoning in technical documents, now public.
-
Gemini 4 Argon (general availability) — Google
Google's new frontier model, broadly available in the Gemini app (AI Pro/Ultra) and API.
-
Muse Wonder — Meta
Meta's new closed models aimed at agentic work, via the Meta Model API and Muse Code.
-
Gemini agent for work — Google Cloud
A single universal work agent inside Gmail, Docs, Sheets and more (announced at Gemini at Work).
-
GPT-6 in ChatGPT — OpenAI
GPT-6 began rolling out in ChatGPT for paid users, with Free and Go following a day later.
-
Claude Haiku 5.5 — Anthropic
Anthropic's smallest, fastest Claude 5.5 model for high-volume and sub-agent work.
-
Free HD video in Google Vids (Gemini Omni) — Google
Anyone with a Google account can generate 1080p AI video clips in Google Vids at no cost.
-
P-Video 2 Pro on Scenario — Pruna AI / Scenario
Text/image-to-video with first/last-frame control and generated audio.
-
Boreal video model — Creatify
AI video model for ads at about one cent per second of video.
-
Fotor Agent — Fotor
Agent that builds editable motion graphics and long-form videos from prompts.