Microsoft Phi-4 Mini Brings AI to Smartphones
Microsoft launches Phi-4 Mini, its smallest language model designed to run entirely on-device on modern smartphones.
📂 LLM News
Latest articles in LLM News
Baidu Launches ERNIE 5.1 at Just 6% Training Cost
Baidu releases ERNIE 5.1 with 'multi-dimensional elastic pre-training,' achieving top benchmark results at a fraction of…
Baidu ERNIE 5.1 Tops China's Search AI Rankings
Baidu launches ERNIE 5.1 with just 6% of typical pretraining costs, ranking 1st in China and 4th globally on LMArena's s…
Teaching Claude Why: Anthropic's Alignment Shift
Anthropic adopts a new alignment philosophy for Claude, focusing on teaching the AI 'why' behind rules rather than just …
OpenAI Launches 3 Realtime Audio Models for Voice AI
OpenAI expands its Realtime API with GPT-Realtime-2, GPT-Realtime-Translate, and GPT-Realtime-Whisper for live voice app…
Claude 4 Opus Shatters Scientific Reasoning Benchmarks
Anthropic's Claude 4 Opus achieves record scores across multiple scientific reasoning benchmarks, outperforming GPT-4o a…
Sarvam AI Launches Open Hindi-English Bilingual LLM
Indian startup Sarvam AI releases an open-source bilingual language model optimized for Hindi and English, targeting 600…
Snowflake Arctic 2 Takes On GPT-4 in SQL Tasks
Snowflake launches Arctic 2, an open-weight enterprise LLM designed to rival GPT-4 on SQL generation and structured data…
Claude Gets 'Dreaming' to Learn From Mistakes
Anthropic adds Dreaming, Outcomes, and Multiagent Orchestration to Claude Managed Agents, enabling AI to reflect and imp…
Anthropic's Claude Can Now 'Dream' to Self-Improve
Anthropic unveils Dreaming, a feature letting Claude agents consolidate memories during idle time, reportedly boosting p…
ZAYA1-8B Matches DeepSeek-R1 on Math With Just 760M Active Params
A new 8B MoE model called ZAYA1-8B achieves DeepSeek-R1-level math performance while activating only 760M of its 8B para…
Kakao Brain Launches Open-Source Korean Multimodal AI
Kakao Brain releases a powerful open-source multimodal model optimized for Korean, directly challenging GPT-4 Vision's d…