Two-Year Anniversary of InstructGPT Release

ByteDance Seed's Latest Reinforcement Learning Recipe POLARIS Open Sourced, 4B Model's Mathematical Reasoning Approaches 235B Performance

Recently, the ByteDance Seed team collaborated with the University of Hong Kong and Fudan University to introduce an innovative reinforcement learning training method called POLARIS. This method successfully enhances the mathematical reasoning capabilities of small models to levels comparable to those of large models through a carefully designed Scaling RL strategy, offering a new approach for optimizing small models in the field of artificial intelligence. Experimental results show that the 4 billion parameter open-source model Qwen3-4B trained using POLARIS achieved remarkable performance on AIME25 and AIME24 mathematical tests.

Mistral Launches Voxtral: The Dawn of a New Era for Open-Source AI Audio Models!

Mistral releases open-source audio model Voxtral, offering cost-effective AI solutions with multilingual support. Three versions available, starting at $0.001/min. Outperforms Whisper at lower cost.....

MiniMax Valued Over 4 Billion USD, Backed by Shanghai State Capital, Joins the 3 Billion USD Large Model Club

Chinese AI firm MiniMax raised $300M, reaching a $4B valuation. Backed by Shanghai state capital, it's now one of China's two $3B+ LLM companies. Founded by ex-SenseTime executives, with prior investments from Alibaba and Tencent, it's reportedly preparing for a Hong Kong IPO.....

A Daily: Moonlight Open-Sources Large Model Kimi K2; Zhiyuan Fully Open-Sources RoboBrain 2.0; Tongyi Qianwen Launches Qwen Chat Desktop Client

Moon's dark side opens trillion-parameter Kimi K2 model; RoboBrain2.0 enhances robot cognition; Alibaba's Qwen adds image generation; IndexTTS2 revolutionizes voice cloning; HuggingFace's Reachy Mini sells well; Meta enables real-time video generation; PixVerse adds multi-keyframe; Tesla Grok supports AMD only; OpenAI delays open-source release; Liquid AI's LFM2 boosts edge AI; AI 'time travel' trend goes viral.....

OpenAI Delays First Open-Source Large Model Release, Ensuring Safety Becomes Top Priority

OpenAI announced the postponement of its first open-source large model release, with CEO Sam Altman stating that more time is needed for safety testing and risk assessment. This new model, which has performance comparable to o3-mini, may be named 'Open Model,' but the extent of its openness remains unclear. Research Vice President Aidan Clark emphasized that the company maintains strict standards for open source, as the model cannot be recalled once released. Although the delay disappointed some users, OpenAI believes ensuring safety and taking a responsible approach is more important. This decision will shape the future of models.

OpenAI Postpones Open-Source Large Model Release, Prioritizes Safety Testing

OpenAI announced the postponement of the open-source large model release. CEO Sam Altman stated that more time is needed for safety testing. The model was originally scheduled to be released this week but is now delayed until next week to ensure its safety and reliability. Altman emphasized that once the model is released, it cannot be recalled and must be handled with caution. This is OpenAI's first attempt to release a downloadable self-running model, aimed at providing powerful tools for researchers and small businesses. Although the delay is disappointing, the community generally understands the importance of safety testing and believes it is crucial for the AI ecosystem.

Major Release! Moonshot Introduces Open-Source Large Model Kimi K2 with Trillion Parameters

MoonDark releases trillion-parameter open-source model Kimi K2, featuring MoE with 32B active params. Excels in coding/math, supports tool use & code execution. Uses MuonClip optimizer, offers API & base model. Key competition to closed-source AI, available via official platforms.....

Mistral AI Releases Devstral2507: Designed for Code-Centric Language Modeling

Mistral AI launched the Devstral2507 series with two AI models: the open-source Devstral Small1.1 (24 billion parameters, SWE-Bench score of 53.6%) and the enterprise version Devstral Medium2507 (score of 61.6%). Small1.1 supports a 128k context window and local deployment, while Medium2507 outperforms some commercial models. Both are optimized for code reasoning and program synthesis, and support integration with agent frameworks.

City Commercial Banks Are Launching a Trend of Large Model Bidding, with Million-Level Investments Becoming a New Industry Opportunity!

Personification of Large AI Models: Grok 4 and Empathy with Musk?