Releases

DeepSeek upgraded its coding API to the Coder V2 generation.

Microsoft releases Florence-2, a unified prompt-based vision model for captioning, grounding, detection and segmentation.

DeepSeek published its first open theorem-proving model for Lean 4.

Mistral AI released Codestral, its first generative model specialized for code, under the Mistral AI Non-Production License.

Cohere Labs released the open Aya 23 35B multilingual model spanning 23 languages.

Cohere Labs released the open Aya 23 8B multilingual model spanning 23 languages.

IBM released Granite Code models from 3B to 34B parameters under a permissive open-source license.

Microsoft introduces Phi-3 Vision, a 4.2-billion-parameter multimodal model for image, chart, graph and table reasoning.

Microsoft expands Phi-3 with 7B Small and 14B Medium open models.

Microsoft releases Phi-3 Medium for higher-capability small-model workloads.

Microsoft introduces Phi Silica, an on-device small language model optimized for Windows neural processing units.

DeepSeek upgraded deepseek-chat with improved instruction following, JSON output and RAG performance.

Baidu introduced its Autonomous Driving Foundation Model for Level 4 driving systems.

Grok-1.5 rolled out to Grok users on X.

Google introduces Imagen 3 with improved photorealism, prompt understanding and text rendering.

Google DeepMind introduces Veo, a high-definition generative video model with cinematic controls.

Google releases PaliGemma, an open vision-language model for image-text tasks.

Google expands Gemini 1.5 Pro to a two-million-token context window in private preview.

Google introduces Gemini 1.5 Flash, a fast and efficient multimodal model with a one-million-token context window.

OpenAI released GPT-4o, an omni model designed for real-time reasoning across audio, vision, and text.

Google DeepMind and Isomorphic Labs introduce AlphaFold 3 for predicting structures and interactions across biomolecules.

DeepSeek published DeepSeek-V2 base and chat weights with 128K context.

DeepSeek introduced its efficient second-generation mixture-of-experts architecture.

Amazon released its second-generation multilingual text embedding model through Bedrock.

Microsoft releases Phi-3 Mini, a 3.8-billion-parameter model designed for capable language and reasoning on constrained devices.

Meta and World Resources Institute released global high-resolution canopy-height maps built with DINOv2.

Meta released the Llama 3-based update to its open safeguard model.

Meta released the 8B Llama 3 pretrained and instruction-tuned models.

Meta released the 70B Llama 3 pretrained and instruction-tuned models.

Hugging Face released IDEFICS2, an 8B visual-language model with stronger OCR, document, and visual reasoning capabilities.

Mistral AI released Mixtral 8x22B, a larger open sparse mixture-of-experts model.

xAI previewed its first multimodal model, adding understanding of documents, diagrams, charts, screenshots and photographs.

Google releases CodeGemma variants for code completion, code generation and instruction following.

Databricks optimized DBRX inference and made the model available through Mosaic AI Model Serving.

Cohere introduced Command R+, its higher-capacity enterprise model for advanced RAG, multilingual generation and tools.

xAI announced Grok-1.5 with improved reasoning and a 128,000-token context window.

Databricks released DBRX, a 132-billion-total-parameter mixture-of-experts model with 36 billion active parameters under an open license.

Cohere Labs made Command R weights available for research and evaluation.

Cohere released Command R, a 128K-context model optimized for production RAG, citations and tool use.

NVIDIA announced a general-purpose foundation model initiative for humanoid robots.

xAI released the 314B-parameter Grok-1 base weights and Mixture-of-Experts architecture under Apache 2.0.

Anthropic released Claude 3 Haiku as the fastest and least expensive model in the Claude 3 family.

Amazon researchers released pretrained Chronos-T5 time-series foundation models and code.

DeepSeek released 1.3B and 7B base and chat checkpoints.

Claude 3 Opus and Claude 3 Sonnet became available on Claude.ai and the generally available Claude API.

Anthropic announced Claude 3 Opus, Sonnet, and Haiku, adding multimodal image understanding and a three-tier capability lineup.

BigCode released StarCoder2 models at 3B, 7B, and 15B scales in collaboration with ServiceNow and NVIDIA.

Mistral AI introduced Mistral Small as its efficient commercial model tier.

Mistral AI launched its first flagship commercial model, Mistral Large.

Google releases Gemma 2B and 7B open models with weights and responsible AI tooling.

Meta released V-JEPA, a video model trained by predicting missing representations rather than pixels.

Google DeepMind introduces Gemini 1.5 Pro with a mixture-of-experts architecture and a one-million-token context window.

OpenAI introduced Sora, a text-to-video model, as a research preview for red teamers and selected creators.

Cohere For AI released Aya 101, an open multilingual instruction-following model covering 101 languages.

Gemini 1.0 Ultra becomes available to consumers through the new Gemini Advanced subscription.

Base, instruction and reinforcement-learning DeepSeekMath 7B checkpoints were published.

Meta released a 70-billion-parameter Code Llama model for code generation and instruction following.

DeepSeek released an updated 7B coding model that later served as the initialization for DeepSeekMath.

OpenAI released text-embedding-3-small and text-embedding-3-large with improved performance and new dimension controls.

Oracle made Meta Llama 2 available as a managed model in OCI Generative AI.

Oracle made Cohere generation and embedding models available through its managed service.

Google makes Imagen 2 available to Cloud customers through Vertex AI for enterprise image generation.

Microsoft releases Phi-2, a 2.7-billion-parameter small language model with strong reasoning and language performance.

Mistral AI released Mixtral 8x7B, an Apache-2.0 sparse mixture-of-experts model.

Meta released Llama Guard, a safeguard model for classifying conversational AI inputs and outputs.

Gemini 1.0 Nano launches for efficient on-device AI, initially on Pixel hardware.

Gemini 1.0 Pro launches in Bard and for developers as the balanced model in the initial family.

Google DeepMind launches Gemini 1.0 in Ultra, Pro and Nano sizes as its first natively multimodal foundation-model family.

Gemini 1.0 Ultra is introduced as the largest and most capable initial Gemini model.

Meta released a low-latency streaming speech translation research model.

Meta released a research model that preserves vocal style and prosody in translated speech.

Meta released a more capable second version of SeamlessM4T.

DeepSeek LLM 67B Base became publicly available as part of DeepSeek's first general LLM release.

DeepSeek published its first general-purpose 7B and 67B open language models.

DeepSeek LLM 7B Base became publicly available as part of DeepSeek's first general LLM release.

DeepSeek LLM 67B Chat became publicly available as part of DeepSeek's first general LLM release.

DeepSeek LLM 7B Chat became publicly available as part of DeepSeek's first general LLM release.

Perplexity introduced PPLX 7B Online and PPLX 70B Online, language models grounded with current web search.

Amazon released a text-to-image generation and editing model through Bedrock.

Amazon released a multimodal embedding model mapping text and images into a shared space.

Anthropic released Claude 2.1 with a 200K-token context window, lower hallucination rates, system prompts, and beta tool use.

Google DeepMind and YouTube introduce Lyria alongside experimental music-generation tools Dream Track and Music AI Tools.

OpenAI announced GPT-4 Turbo with a 128K context window, newer knowledge, and lower API pricing.

xAI announced Grok and its first public model after two months of training, with real-time knowledge through X.

Base and instruction-tuned DeepSeek Coder checkpoints were published for code completion and generation.

DeepSeek released its first open code-model family, spanning 1B to 33B parameters and trained on two trillion tokens.

Cohere introduced its third-generation embedding models with stronger multilingual retrieval and compression options.

Moonshot exposed a 128K-context text model through its API.

Moonshot exposed a 32K-context text model through its API.

Moonshot exposed an 8K-context text model through its API.

Cohere released third-generation English and multilingual reranking models for enterprise search.

Hugging Face H4 released an improved Zephyr 7B beta model trained on expanded public and synthetic preference data.

OpenAI made DALL-E 3 available to ChatGPT Plus and Enterprise users.

Baidu launched ERNIE 4.0 with upgraded comprehension, reasoning, generation and memory.

Perplexity released open-weight PPLX 7B Chat and PPLX 70B Chat models.

IBM made its first 13B Granite enterprise language models available through watsonx.ai.

IBM released the first enterprise-focused Granite language and code models in watsonx.

Hugging Face H4 released an aligned 7B chat model trained with distilled direct preference optimization.

Amazon released Titan Text Express for general text-generation workloads through Bedrock.

Amazon released its lower-cost Titan text-generation model through Bedrock.