Meta introduces Muse Spark with a restricted developer preview
The first model from Meta Superintelligence Labs began in Meta AI.
Capvise Media / Stories
AI
The systems learning, generating and acting around us.
90 storiesBrowse by month
The AI model gets the credit. What is doing the rest of the work?
Sessions, tools, sandboxes and evaluation loops help turn one response into a longer job.
Gemma 4 arrives under Apache 2.0 in four model sizes
The family spans edge models, a mixture-of-experts model and a dense model.
Gemini 3.1 Flash Live updates Google’s real-time voice stack
Search Live expansion accompanied developer-preview access.
Lyria 3 Pro extends generated music to three minutes
Google expanded access through developer tools and creative products.
Mistral’s first Voxtral TTS model generates speech in nine languages
Studio and API access target custom voice applications.
GPT-5.4 mini and nano add smaller tiers for repeated tasks
Nano remains an API-only option at launch.
Mistral Forge offers model training around internal company knowledge
The system spans pretraining, post-training and reinforcement learning.
Mistral Small 4 unifies reasoning, image input and agent coding
The 119B-parameter model activates a smaller subset for each token.
Leanstral targets proof engineering in Lean 4 repositories
Formal verification checks a specification rather than the intention behind it.
GPT-5.4 brings computer use into a general purpose model
The March release joined reasoning, coding and work across applications.
Gemini 3.1 Flash-Lite enters preview at $0.25 per million input tokens
Thinking controls let developers vary effort for high-volume tasks.