Releases

Meta released a unified multimodal model for separating sounds from complex mixtures using natural prompts.

NVIDIA introduced Nemotron 3 Nano, Super and Ultra open models and companion reinforcement-learning and evaluation tools.

Cohere launched Rerank 4, its fourth-generation retrieval model family in Pro and Fast tiers.

OpenAI released GPT-5.2 with improvements for professional knowledge work and long-context tasks.

Mistral AI released Devstral 2, its second-generation open code-agent model.

Mistral AI released the Mistral 3 generation, including open-weight Mistral Large 3 and Ministral 3 models in 3B, 8B and 14B sizes.

AWS released a unified embedding model for text, documents, images, video and audio.

AWS released its second-generation speech-to-speech foundation model.

AWS introduced a unified multimodal reasoning and image-generation model in preview.

AWS introduced its most intelligent Nova 2 model for complex multistep tasks in preview.

AWS released a fast multimodal reasoning model with extended thinking, built-in tools and a one-million-token context window.

DeepSeek offered a temporary high-compute reasoning endpoint until December 15.

DeepSeek upgraded chat and reasoner services to the production V3.2 model.

DeepSeek released its self-verifiable mathematical reasoning model.

Anthropic released Claude Opus 4.5 for coding, agents, computer use, research, and professional work.

Google launches Gemini 3 Pro Image, branded Nano Banana Pro, for professional image generation and editing.

xAI released Grok 4.1 Fast and an Agent Tools API with X search, web search, code execution and collections search.

Meta released a model for estimating human body pose and shape from a single image.

Meta released a model for reconstructing 3D objects and scenes from a single image.

Meta released SAM 3 for text-, exemplar- and visual-prompted detection, segmentation and tracking in images and video.

Google launches Gemini 3 Pro for advanced multimodal reasoning, coding and agentic work.

xAI released Grok 4.1 across grok.com, X, iOS and Android with improvements to creative, emotional and collaborative interactions.

OpenAI released GPT-5.1 for developers with improved coding, steerability, and agentic performance.

Baidu released a sparse multimodal reasoning model activating only 3B parameters.

Meta released speech-recognition models covering more than 1,600 languages plus an open corpus for 350 underserved languages.

Moonshot released the reasoning-focused K2 Thinking model for long-horizon agentic tasks.

DeepSeek published its optical-compression model for OCR and document understanding.

Amazon released a universal time-series foundation model for univariate, multivariate and covariate-informed forecasting.

Baidu released a compact multilingual vision-language model for document parsing.

Google launches Veo 3.1 with richer audio, stronger narrative control and reference-image guidance.

Anthropic released Claude Haiku 4.5, offering near-frontier coding performance at lower cost and latency.

Microsoft AI announces MAI-Image-1, its first image-generation model developed entirely in-house.

Google releases Gemini 2.5 Computer Use in preview for agents that interact with graphical user interfaces.

IBM launched hybrid Mamba/transformer Granite 4.0 models optimized for efficient agentic workloads.

Moonshot published a 48B Kimi Linear checkpoint.

OpenAI released Sora 2, a video and audio generation model with improved physical realism and control.

DeepSeek released the experimental V3.2 model with sparse attention.

NVIDIA released GR00T N1.6, Cosmos Reason NIM and new Newton physics integration for Isaac Lab.

Anthropic released Claude Sonnet 4.5 with major gains in coding, computer use, agents, reasoning, and mathematics.

Google DeepMind introduces Gemini Robotics 1.5 and an embodied-reasoning model for more capable robot agents.

Gemini 2.5 Flash-Lite becomes generally available for high-volume, latency-sensitive applications.

DeepSeek improved language consistency and its code and search agents.

xAI released Grok 4 Fast with a 2M-token context window, unified reasoning modes and lower-cost agentic search.

Baidu introduced a sparse-attention inference update for faster long-context performance.

OCI Generative AI added beta access to OpenAI's gpt-oss open-weight models.

Moonshot released a K2 checkpoint with improved coding, instruction following and long-context performance.

Google releases EmbeddingGemma, an open multilingual embedding model optimized for on-device retrieval.

Cohere released its first dedicated machine-translation model, supporting 23 languages and private deployments.

xAI released a speedy and economical reasoning model optimized for agentic coding.

Microsoft AI releases MAI-Voice-1, its first in-house expressive speech-generation model, through Copilot experiences.