Blog

Google DeepMind releases Gemma, an open-source AI model family derived from Gemini research, featuring 2B and 7B parameter variants optimized for advanced reasoning and commercial use.

Google DeepMind releases Gemini 1.5 Pro with a groundbreaking 1 million token context window, setting a new standard for enterprise-scale AI applications.

Google DeepMind releases Gemini 1.0 Ultra, the most powerful multimodal AI model yet, outperforming GPT-4 on 30 of 32 benchmarks.

Stability AI releases StableLM 2 with 1.6B and 12B parameters, trained on 2 trillion tokens, delivering competitive performance against larger models.

BigCode and ServiceNow release StarCoder 2, a family of open-source code generation models with 16K context and support for 600+ programming languages.

Discover how DeepSeekMath-7B uses GRPO and a massive 120B token corpus to rival industry giants in mathematical reasoning.

DeepSeek-MoE marks a historical milestone in open-source AI, proving that sparse Mixture-of-Experts models can match dense giants with a fraction of the compute.

Korean startup Upstage releases SOLAR 10.7B, a groundbreaking open-source model that topped the HuggingFace Open LLM Leaderboard upon release.

Mistral AI's groundbreaking 46.7B parameter MoE model delivers GPT-3.5-level performance with just 12.9B active parameters.