AI Daily · Nov 21, 2026: Google ships Gemini 4 Hydrogen, its third Gemini model in eight days
Google launched speed-focused Gemini 4 Hydrogen in the Gemini app and Search, its third Gemini model in eight days, while Thinking Machines' 26B Inkling 2 Mini now runs locally via Ollama at up to 210 tok/s decode.
Pick a date
-
Google launches Gemini 4 Hydrogen in Gemini app and Search
Gemini 4 Hydrogen is now live in the Gemini app and Google Search, positioned for speed and everyday work across diverse agentic workflows.
-
Google releases Neon, its third Gemini model in eight days
Google released Neon, described as its third Gemini model in eight days, alongside an expansion of Tasks in the Gemini app.
-
Inkling 2 Mini from Thinking Machines lands on Ollama pre-release
Inkling 2 Mini (A3B/26B, 5.4B active parameters) is available in Ollama 0.40.17 pre-release and runs on Macs with 32GB+ memory; each token passes through its middle layers three times for deeper reasoning, targeting coding, computer use and visual design. Hugging Face reports support in transformers, llama.cpp and MLX.
-
Inkling 2 Mini runs locally at 210 tok/sec on M5 Ultra
Exo Labs reports the 26B-parameter Inkling 2 Mini decodes at 210 tok/sec and prefills at 3620 tok/sec on a 96GB M5 Ultra Mac Studio, and 125 tok/sec decode on a 48GB M5 Max MacBook Pro; it built a weather app 1.8x faster and 11x cheaper than Claude Sonnet 5.5.
-
HeyGen Decision 1 launches on OpenRouter with 50% off pricing
Decision 1 answers up to 400 questions per call across video, audio, image and text with a calibrated probability per answer in about 100 ms. Launch pricing is 50% off through Dec 1: Turbo at $0.042 per 1M input tokens, Decision 1 at $0.07 per 1M input tokens, output free.
-
Command Code offers free and half-price models including MiniMax M3
Coding agent Command Code lists Glyph Cluster and Flint as free models and MiniMax M3 at 50% off for users.
-
Google brings Gemini 3.5 Transcribe voice editing to the Gemini API
Google is opening Gemini 3.5 Transcribe's voice editing to the Gemini API, letting developers build experiences where users edit, rephrase and translate existing text by voice; the same tech powers Rambler on Gboard and Gemini on macOS.
-
OpenAI opens ChatGPT agent publishing to non-US businesses
Ecommerce and online service businesses based outside the U.S. can now claim their business and publish agents in ChatGPT.
-
OpenAI publishes research on detecting and disrupting model distillation
OpenAI shared findings on how distillation can transfer powerful capabilities without their safeguards, and on methods to detect and disrupt coordinated distillation attempts.