LLMs & Foundation Models
Breakthroughs in large language models, reasoning architectures, context windows, and multimodal AI systems.
Published Reporting (5 Stories)
Sorted by recency
DeepSeek R1 Open-Weights Model Matches Frontier Reasoning Benchmarks on Mathematics and Coding
DeepSeek has publicly released weights for DeepSeek-R1, an open reasoning model trained via large-scale reinforcement learning that matches proprietary frontier models on competitive coding and Olympiad math.
Claude 3.7 Sonnet Introduces Hybrid Reasoning Architecture with Dynamic Thinking Budget Control
Anthropic has unveiled Claude 3.7 Sonnet, the first hybrid frontier foundation model that combines instantaneous conversational speed with adjustable, extended chain-of-thought deliberation for complex coding tasks.
Google Gemini 2.0 Flash Delivers Real-Time Multimodal Streaming and Native Audio-Video Latency Under 250ms
Google has released Gemini 2.0 Flash into general availability, featuring native real-time audio and vision streaming capabilities with sub-250ms latency for ambient computing applications.
Llama 3.3 70B Delivers 405B-Grade Performance with 80% Reduction in Memory Footprint
Meta AI has released Llama 3.3 70B, an optimized open-weights language model that matches the benchmark performance of the previous 405B parameter flagship while fitting entirely onto a single commercial GPU node.
OpenAI o3-Mini Reasoning Model Achieves Gold-Medal Competitive Math and Competitive Programming Standards
OpenAI has released o3-mini, a cost-effective reasoning model featuring low, medium, and high deliberation effort settings tailored for high-accuracy STEM, science, and coding workflows.