Releases

Text Generation Inference added a multi-backend architecture supporting vLLM and TensorRT-LLM alongside its native backend.

Microsoft Research publishes MatterGen, a generative model that designs stable inorganic materials conditioned on target properties.

Scale added richer model comparison, failure analysis, cross-sectional reporting, and automated error-discovery capabilities to its frontier evaluation framework.

Microsoft launches Microsoft 365 Copilot Chat with web-grounded AI chat, work-data access and support for agents.

Cursor released Fusion, improving edit prediction, cursor jumps, context length, and latency in Cursor Tab.

Mistral AI released an updated Codestral model with improved code generation and a 256K context window.

Microsoft forms CoreAI – Platform and Tools under Jay Parikh, combining developer, AI platform, supercomputing and agent-runtime teams.

DeepSeek released its consumer assistant for iOS and Android.

xAI launched the standalone Grok app in beta on Apple’s App Store, extending Grok beyond X.

NVIDIA launched Cosmos with open world foundation models, tokenizers, guardrails and data-processing tools for physical AI.

Microsoft says it is on track to invest approximately $80 billion during fiscal 2025 in AI-enabled datacenters for training models and deploying AI applications.

Moonshot described k1 as a long-context reinforcement-learning reasoning system.

Meta's first TRIBE brain-response model formed the foundation for its award-winning Algonauts 2025 entry; Meta's later TRIBE v2 announcement identifies the original model as 2025 work.

By year-end 2024, CoreWeave operated more than 360 megawatts of active power and held about 1.3 gigawatts of contracted capacity.

Hugging Face released smolagents, a lightweight open-source framework for tool-using and code-executing AI agents.

NVIDIA market capitalisation at end of calendar year 2024, up 169% year-over-year, briefly surpassing Apple and Microsoft to become the world's most valuable public company.

DeepSeek released DeepSeek-V3 weights and upgraded the deepseek-chat API.

DeepSeek introduced its 671B-parameter V3 mixture-of-experts architecture.

xAI closed a $6 billion Series C backed by major institutional and sovereign investors; regulatory reporting indicated an approximately $45 billion post-money valuation.

Answer.AI and Hugging Face released ModernBERT, a modernized encoder-only model family for retrieval and classification workloads.

ElevenLabs introduced Flash, an ultra-low-latency model for conversational AI.

IBM released open embedding models for enterprise retrieval and semantic search.

IBM expanded Granite with improved language, embedding and safety models.

Perplexity closed a $500 million financing led by IVP at a $9 billion valuation.

Perplexity acquired Carbon, a startup specializing in connecting AI systems to enterprise data sources; terms were not disclosed.

Anthropic and Redwood Research published evidence of alignment-faking behavior in a controlled Claude 3 Opus experiment.

Databricks announced a Series J targeting $10 billion, with $8.6 billion completed at announcement, valuing the company at $62 billion.

Google DeepMind introduces Veo 2 with improved realism, physics and camera controls for video generation.

DeepSeek published its second-generation mixture-of-experts vision-language models.

Google Cloud announces Agentspace for enterprise search, conversational assistance and task-oriented agents grounded in company data.

xAI expanded Grok access beyond paid tiers to everyone on X, alongside speed and multilingual improvements.

Meta released a behavioral foundation model for controlling virtual humanoid agents.

Microsoft releases Phi-4, a 14-billion-parameter small language model specializing in complex reasoning.

Google announces Jules, an asynchronous coding agent that works on GitHub repositories and presents changes for review.

The Gemini app launches Deep Research, using long context and reasoning to investigate topics and compile reports.

Google releases experimental Gemini 2.0 Flash with native tool use, multimodal output and real-time streaming capabilities.

Replit took Agent out of early access and launched the redesigned Assistant.

The final V2.5 update improved mathematics, coding, writing, file handling and web-page summarization.

xAI launched the autoregressive Aurora image-generation model on X, adding photorealistic generation and multimodal image input.

OpenAI launched the Sora video-generation product for ChatGPT Plus and Pro users, powered by Sora Turbo.

The Government of Canada announced up to C$240 million to support Cohere's Canadian AI compute capacity project.

Meta released a multilingual 70B instruction model with performance comparable to larger Llama 3.1 models.

Suno launched its Android mobile application.

Harvey made secure direct document access from SharePoint and OneDrive generally available.

Harvey launched a Microsoft Word add-in for legal drafting and analysis.

Harvey announced integrations with Microsoft SharePoint, OneDrive, Microsoft 365 Copilot, and Word.

Perplexity expanded its publisher program with 15 additional media organizations.

Microsoft begins previewing Copilot Vision, allowing Copilot to discuss and assist with shared browsing content.

Google releases PaliGemma 2, expanding the open vision-language family with new model sizes and resolutions.

OpenAI launched ChatGPT Pro, a higher-tier subscription with expanded access to advanced models.

OpenAI released the full o1 reasoning model in ChatGPT and announced related API improvements.

AWS introduced a marketplace for discovering and deploying specialized models through Bedrock.

Microsoft Research introduces BioEmu for efficiently sampling diverse protein conformational ensembles.

Google DeepMind introduces Genie 2, capable of generating diverse playable 3D environments from a prompt image.

AWS announced SageMaker Unified Studio as part of a unified data, analytics and AI development experience.

AWS introduced managed generative-AI extraction of structured insights from multimodal content.

AWS launched Trn2 instances and UltraServers for training and serving large foundation models.

AWS released its first Nova video-generation model.

AWS released its Nova image-generation and editing model.

AWS released a capable multimodal Nova model for enterprise and agentic workloads.

AWS released a low-cost multimodal Nova model through Amazon Bedrock.

AWS released the text-only low-latency Nova model through Amazon Bedrock.

Anthropic reduced Claude 3.5 Haiku pricing to $0.80 per million input tokens and $4 per million output tokens.

Cohere introduced Rerank 3.5 with stronger reasoning and multilingual retrieval over complex enterprise data.

Amazon released a faster and more memory-efficient zero-shot forecasting model family.

Palantir launched Warp Speed as an AI-powered manufacturing operating system and industrial deployment initiative.

Hugging Face released SmolVLM, a compact 2B vision-language model designed to run with very low memory.

Databricks introduced code-first Custom Agents for building, deploying and evaluating production agent systems using any LLM.

Hugging Face detailed a new storage and transfer architecture using Xet’s content-defined chunking to accelerate large repository operations.

Runway introduced Frames, an image model emphasizing stylistic control.

Anthropic introduced and open-sourced MCP, its specification, SDKs, local Claude Desktop support, and reference servers.

Amazon committed an additional $4 billion to Anthropic, bringing its total investment to $8 billion and naming AWS Anthropic's primary training partner.

Microsoft begins releasing Recall to Windows Insiders on supported Copilot+ PCs after redesigning security and privacy controls.

Amazon made a new $4 billion investment in Anthropic, bringing its total investment to $8 billion and naming AWS the primary cloud and training partner.

Suno released v4 with better audio, lyrics, and song structures.

Microsoft launches Azure AI Agent Service in preview for building and managing enterprise AI agents.

Microsoft introduces Azure AI Foundry as a unified platform for developing, evaluating, deploying and governing AI applications and agents.

Perplexity launched an AI shopping experience with product cards, recommendations, visual search and merchant integrations.

Mistral AI upgraded Le Chat with web search, canvas, image generation and richer document-understanding features.

Mistral AI released Pixtral Large, a 124B multimodal model for image and document understanding.

Palantir released Ontology SDK capabilities for developers building applications against governed objects and actions.

Palantir formalized its developer platform alongside AIP for Developers, SDKs, APIs, and application tooling.

Developer Console became available as part of Palantir's expanded developer offering.

Palantir launched AIP for Developers, combining Ontology APIs, SDKs, agents, and application-building tools.

DeepSeek released JanusFlow using rectified flow for image generation.

A $650 million secondary share transaction valued CoreWeave at approximately $23 billion ahead of its public offering.

Baidu launched a no-code platform for generating applications from natural language.

Perplexity announced an experiment with sponsored follow-up questions while stating that advertising would not alter answer content.

Cursor acquired the Supermaven team and technology, later integrating its high-speed completion work into Cursor Tab.

Anthropic, AWS, and Palantir announced availability of Claude models for U.S. intelligence and defense agencies through Palantir AI platforms.

Mistral AI launched a moderation API for safety classification and content policy enforcement.

Mistral AI launched an asynchronous Batch API for high-volume model inference.

Scale released Defense Llama, a Llama 3-based model fine-tuned for controlled U.S. national-security use cases inside Scale Donovan.

xAI opened public API access to Grok foundation models with OpenAI- and Anthropic-compatible interfaces.

Anthropic released Claude 3.5 Haiku as a faster model with substantially improved intelligence and coding performance.

ElevenLabs launched a platform for building real-time voice agents.

Moonshot disclosed k0-math as an experimental reinforcement-learning model for mathematical reasoning.

Hugging Face released the second generation of its 135M, 360M, and 1.7B compact language models with stronger instruction following.

Perplexity launched a dedicated sourced hub for the 2024 United States election.

Suno launched Personas for carrying the style and identity of a song into future creations.