DeepSeek releases the open-weight reasoning model R1
DeepSeek publishes R1, R1-Zero, six distilled models, and its reinforcement-learning recipe, pushing open reasoning models into the frontier conversation.
SEARCH
Search English reports
English reports are listed here. AI-generated drafts are clearly labeled until an editor reviews them.
AI-generated · reviewedImpact 99/100
ByteDance Seed launches the Doubao Realtime Voice Model
ByteDance Seed launches the Doubao Realtime Voice Model and makes it available in the Doubao app.
DeepSeek launches a free consumer app before R1
DeepSeek puts V3, web search, a reasoning mode, and file analysis into a free mobile app five days before the R1 release.
Mistral AI releases Mistral Small 3.0
Mistral AI releases a third-generation lightweight foundation model optimized for performance on limited hardware.
DeepSeek-V3 brings an efficiency-first open model to the frontier race
DeepSeek releases a 671B MoE model with 37B active parameters, open weights, low API prices, and unusually detailed efficiency disclosures.
Google DeepMind releases Gemini 2.0 Flash Thinking
Google DeepMind releases a derivative of the 2.0 Flash architecture with explicit reasoning for logic-heavy tasks.
Technology Innovation Institute releases Falcon 3
The Technology Innovation Institute releases a new foundation-model family covering several smaller parameter scales.
Microsoft releases Phi-4
Microsoft releases its fourth-generation small-parameter model, using substantial synthetic data to strengthen basic reasoning.
Google DeepMind releases Gemini 2.0 Flash
Google DeepMind releases a second-generation lightweight foundation model with faster responses and improved concurrent multitask handling.
OpenAI launches Sora as a public video product
OpenAI moves Sora from research preview to a standalone video-generation product for ChatGPT Plus and Pro users.
xAI releases the Aurora image model
xAI released Aurora, an in-house autoregressive mixture-of-experts image model trained on interleaved text and image data, through Grok on X.
LG AI Research releases EXAONE 3.5
LG AI Research releases a medium-scale model optimized for Korean-English bilingual processing.
Meta releases Llama 3.3
Meta releases a 70-billion-parameter model that consolidates long-context and multilingual capabilities.
OpenAI releases o1-pro
OpenAI releases an advanced o1 variant that uses more compute for longer reasoning to improve accuracy.
Amazon releases Amazon Nova
Amazon AWS releases a new family of multimodal foundation models.
Tencent releases HunyuanVideo inference code and model weights
Tencent released HunyuanVideo, a text-to-video model with more than 13 billion parameters, together with model weights and inference code.
Upstage releases SOLAR Pro
Upstage releases a 22-billion-parameter model optimized for document understanding and long-context tasks.
DeepSeek previews its reasoning line before V3 ships
R1-Lite-Preview exposes long-form reasoning on the web and promises an open model and API, showing that R1 was already underway before V3's release.
Allen Institute for AI releases OLMo 2
The Allen Institute for AI releases an open model while keeping its training data and underlying code fully public.
Mistral AI releases Pixtral Large
Mistral AI releases a large multimodal model with stronger understanding of visual features.