Microsoft releases Phi-3
Microsoft releases the third Phi generation, optimized for efficient execution on edge devices.
EVENT ARCHIVE
Events
Browse published English reports with original sources and content provenance.
AI-assisted · reviewed
Meta releases Llama 3
Meta releases its fourth foundation-model generation with more training data and stronger same-scale performance.
Mistral AI releases Mixtral 8x22B
Mistral AI releases an open mixture-of-experts model with a 176-billion-parameter scale.
Microsoft releases WizardLM-2
Microsoft releases a new open model family with strong instruction following and complex-task performance.
Cohere releases Command R+
Cohere releases an enterprise model optimized for retrieval-augmented generation and tool use.
AI21 Labs releases Jamba
AI21 Labs releases a hybrid model combining Mamba and Transformer architectures.
Databricks releases DBRX
Databricks releases an open 132-billion-parameter mixture-of-experts model that raises open-model benchmark performance.
xAI releases Grok-1 weights and architecture
xAI released the 314-billion-parameter Grok-1 base-model weights and network architecture under Apache 2.0.
Anthropic releases Claude 3
Anthropic releases the Claude 3 family with Haiku, Sonnet and Opus variants and image understanding.
Google DeepMind releases Gemma
Google DeepMind releases a lightweight open model derived from the Gemini technology architecture.
Google DeepMind releases Gemini 1.5 Pro
Google DeepMind releases a model with a new architecture and native support for context windows of up to 1 million tokens.
Gemini 1.0 Ultra launches in Gemini Advanced
Google makes Gemini 1.0 Ultra available through Gemini Advanced, moving the largest Gemini 1.0 model from announcement to a consumer product.
DeepSeekMath introduces the GRPO method later used by R1
DeepSeek releases math models and Group Relative Policy Optimization, creating a direct algorithmic predecessor to R1's reinforcement-learning recipe.
DeepSeek publishes the sparse architecture behind its later models
DeepSeekMoE uses fine-grained experts and shared expert isolation to improve sparse-model efficiency, with 16B weights and training code released.
Upstage releases SOLAR 10.7B
Upstage releases a 10.7-billion-parameter model built with depth up-scaling.
Microsoft releases Phi-2
Microsoft releases a small-parameter model demonstrating the value of high-quality, textbook-style training data.
Mistral AI releases Mixtral 8x7B
Mistral AI releases an open model using an 8x7B mixture-of-experts architecture that balances capability and inference efficiency.
Google releases the Gemini 1.0 family
Google introduces the natively multimodal Gemini 1.0 family—Ultra, Pro and Nano—with Pro and Nano entering products immediately and developer APIs following a week later.
Microsoft releases Orca 2
Microsoft releases the second Orca generation to improve reasoning in small-parameter models.
Nous Research releases Nous Hermes 2
Nous Research releases a conversational model fine-tuned from an open base model by the open-source community.