Releases

Hivemind Commander provided mission planning and control within the Enterprise suite.

Hivemind Forge became the simulation and autonomy-development environment within Hivemind Enterprise.

Shield AI launched Hivemind Enterprise as a complete autonomy-development suite.

Shield AI completed a $240 million strategic financing at a $5.3 billion valuation.

Mistral AI launched its first OCR model and Document AI capabilities for high-quality document parsing.

Anthropic upgraded its developer Console with tools for prompt development, evaluation, collaboration, and production deployment.

The U.S. Defense Innovation Unit selected Scale to lead Project Thunderforge with Anduril and Microsoft, integrating AI agents into operational military planning.

Google begins testing AI Mode, a conversational Search experience for complex, multimodal questions and follow-ups.

CoreWeave entered an agreement to acquire Weights & Biases and add experiment tracking, model management, evaluation, and monitoring to its AI cloud platform.

Microsoft introduces Dragon Copilot, combining ambient clinical documentation, dictation and healthcare workflow assistance.

Anthropic raised $3.5 billion led by Lightspeed Venture Partners at a $61.5 billion post-money valuation.

Moonshot released a preview checkpoint for formal mathematical theorem proving.

Moonshot introduced the Kimina-Prover line of open models for formal theorem proving.

IBM completed the HashiCorp acquisition, adding Terraform and Vault to its hybrid-cloud and AI platform.

Anthropic launched a public Transparency Hub describing its safety, governance, security, and responsible-deployment practices.

OpenAI released GPT-4.5 as a research preview in ChatGPT and the API.

ElevenLabs launched transcription powered by Scribe.

ElevenLabs released Scribe, its first major speech-to-text model.

IBM released updated compact time-series models for long-range forecasting.

IBM updated its open safety model for detecting risks in prompts and responses.

IBM released a compact vision-language model for document and chart understanding.

IBM released the 8B Granite 3.2 model with controllable reasoning capabilities.

IBM released the 2B Granite 3.2 model with controllable reasoning capabilities.

Amazon announced a generative-AI version of Alexa with natural conversation, personalization and agentic task completion.

Microsoft releases Phi-4 Multimodal, a 5.6-billion-parameter model jointly processing speech, vision and text.

Microsoft releases Phi-4 Mini, a compact long-context language model with multilingual and function-calling capabilities.

Replit released Agent 2 with real-time application design preview.

IBM agreed to acquire DataStax to add vector data, Astra DB and Langflow capabilities to watsonx.

Perplexity announced a $50 million venture fund to invest in early-stage AI startups.

Anthropic introduced Claude Code as a terminal-based agentic coding tool in limited research preview.

Anthropic released Claude 3.7 Sonnet, its first hybrid reasoning model with standard and extended-thinking modes.

Scale entered a five-year partnership with Qatar’s Ministry of Communications and Information Technology to deploy AI applications, analytics, automation, and workforce programs across government.

Hugging Face released SmolVLM2 at 256M, 500M, and 2.2B parameters with support for video understanding as well as images and text.

xAI unveiled Grok 3 and Grok 3 mini with reasoning modes, DeepSearch and Think capabilities trained on Colossus.

Microsoft Research and Xbox introduce Muse, a generative world and action model trained on gameplay sequences.

Perplexity released R1 1776, an open-weight post-trained version of DeepSeek-R1 intended to reduce censorship while retaining reasoning capability.

South Korea's privacy regulator suspended new downloads while reviewing personal-data protections.

Mistral AI released Mistral Saba, a 24B regional language model optimized for Arabic and languages of the Middle East and South Asia.

Perplexity launched Deep Research, an autonomous mode that performs many searches, evaluates sources and writes cited reports.

Harvey raised $300 million at a $3 billion valuation in a Sequoia-led Series D.

Scale published MASK as part of its safety research program for testing whether models knowingly state beliefs inconsistent with their internal representations.

Scale released FORTRESS to evaluate frontier-model safeguards against dual-use national-security and public-safety risks.

Perplexity released a faster in-house Sonar model built on Llama 3.3 70B and served through Cerebras for Pro search.

Cohere and RBC began co-developing North for Banking for secure, regulated financial-services work.

Cohere introduced Compass as North's multimodal search and discovery backbone for governed enterprise data.

Cohere opened early access to North, an all-in-one secure AI workspace combining models, enterprise search and agents.

A consortium led by xAI submitted an unsolicited $97.4 billion offer for the nonprofit controlling OpenAI; the proposal was rejected and did not proceed.

Anthropic launched the Economic Index to measure how AI is being used across occupations and tasks.

Meta released PARTNR, a benchmark, dataset and planning model for human-robot collaboration on household tasks.

Mistral AI relaunched Le Chat with iOS and Android apps, Pro and Team plans, faster inference, image generation and an Enterprise preview.

Meta researchers introduced a non-invasive deep-learning system for decoding typed sentences from EEG and MEG brain recordings.

Google releases Gemini 2.0 Flash-Lite in public preview as its lowest-cost Gemini 2.0 model.

Google releases an experimental Gemini 2.0 Pro for coding and complex prompts with a two-million-token context window.

Gemini 2.0 Flash becomes generally available through the Gemini API, Google AI Studio and Vertex AI.

Australia directed federal entities to remove and prevent DeepSeek products on government systems and devices.

CoreWeave became the first cloud provider to make rack-scale NVIDIA GB200 NVL72 instances generally available through its Kubernetes service.

xAI released the standalone Grok app for Android.

Anthropic published and tested Constitutional Classifiers, a safeguard approach designed to resist broad classes of jailbreaks.

OpenAI launched deep research, an agentic ChatGPT capability for multi-step web research and synthesis.

Moonshot released Moonlight, a 16B MoE model used to evaluate Muon optimization at scale.

xAI completed the second phase of Colossus 1, doubling the interconnected cluster to roughly 200,000 NVIDIA H100/H200 GPUs.

OpenAI released o3-mini, a cost-efficient reasoning model optimized for science, mathematics, and coding.

ElevenLabs raised $180 million at a $3.3 billion valuation.

Italy's data-protection authority ordered an immediate restriction on DeepSeek's processing of Italian users' personal data.

Mistral AI released Mistral Small 3, a 24B latency-optimized open-weight model.

Google rolls out Gemini 2.0 Flash as the default model in the Gemini app.

CoreWeave detailed Mission Control capabilities for infrastructure validation, proactive node replacement, fleet health, and AI cluster observability.

Perplexity introduced Sonar Reasoning and Sonar Reasoning Pro for search-grounded analytical queries.

Hugging Face launched a unified inference layer that lets developers call models across multiple providers through one API and billing experience.

IBM launched an AI-powered delivery platform for consultants and client transformation work.

DeepSeek released scaled 1B and 7B Janus-Pro checkpoints with improved understanding and image generation.

DeepSeek's assistant reached number one among free iPhone applications in the United States as adoption surged.

DeepSeek temporarily limited new registrations following a large-scale malicious cyberattack.

Hugging Face launched Open R1 to openly reproduce and extend the data, training, and evaluation pipeline behind reasoning models.

Scale and the Center for AI Safety published Humanity’s Last Exam, assembled from nearly 1,000 contributors across more than 500 institutions in 50 countries.

Hugging Face released a 500M vision-language model for low-memory multimodal inference.

Hugging Face released a 256M vision-language model small enough for highly constrained devices while supporting image-and-text tasks.

Perplexity launched an Android assistant able to answer questions and perform actions across supported apps.

OpenAI launched Operator, a browser-using agent capable of interacting with web interfaces.

Databricks completed $10 billion in equity financing and arranged $5.25 billion in debt facilities at a $62 billion equity valuation.

Google makes an additional $1 billion investment in Anthropic, extending its financial and cloud partnership with the AI developer.

Google reportedly made an additional $1 billion investment in Anthropic, deepening its existing investment and cloud relationship.

Modal made isolated programmable Sandboxes generally available for agent-generated code.

Modal launched dedicated infrastructure for secure code-interpreting and agent workloads.

Perplexity launched Sonar and made Sonar Pro generally available through its API platform for cited real-time answers.

Microsoft and OpenAI update their partnership structure while preserving Microsoft rights to OpenAI intellectual property and Azure API exclusivity.

OpenAI, SoftBank, Oracle, and MGX announced Stargate, a new company intended to invest $500 billion in US AI infrastructure over four years.

Moonshot released Kimi k1.5, a multimodal reasoning model trained with reinforcement learning.

Moonshot introduced its open k-series reasoning work through the Kimi k1.5 release.

DeepSeek released the Llama 70B-based distilled R1 reasoning checkpoint.

DeepSeek released the Llama 8B-based distilled R1 reasoning checkpoint.

DeepSeek released the Qwen 14B-based distilled R1 reasoning checkpoint.

DeepSeek released the Qwen 1.5B-based distilled R1 reasoning checkpoint.

DeepSeek released the Qwen 7B-based distilled R1 reasoning checkpoint.

DeepSeek released the Qwen 32B-based distilled R1 reasoning checkpoint.

DeepSeek published R1 under the MIT License and made it available on web and API.

DeepSeek published its pure reinforcement-learning reasoning checkpoint.

DeepSeek introduced its reinforcement-learning reasoning model family and technical report.

Cursor disclosed that annual recurring revenue had surpassed $100 million.

Cursor raised $105 million in Series B funding from Thrive Capital, Andreessen Horowitz, Benchmark, and existing investors at a reported $2.6 billion post-money valuation.