China's AI Model Lands on Nature Cover! DeepSeek Reveals R1 Training Costed Just $2 Million
DeepSeek R1's cover feature in Nature sparks debate over training costs, scientific evaluation, and discourse power.
Models & Benchmarks
Releases, papers, SOTA benchmarks
Filter by category
2025 The Bund Summit: Ant Group and Renmin University Release First Native MoE Diffusion Language Model, Open Source Coming Soon
Ant Group and Renmin University unveil a diffusion language model roadmap, exploring the potential paradigm of combining Mixture-of-Experts with diffusion for language modeling.
European OpenAI Faces Backlash Over Alleged DeepSeek Distillation and Fabrication
A European version of OpenAI is under fire for allegedly distilling data from DeepSeek and fabricating results, leading to a significant credibility crisis.
"Give Us Back GPT-4o!" OpenAI's Push for GPT-5 Sparks Outrage, Prompting Emergency PR
GPT-5 pricing and price wars: API rates seen as pressure on competitors, sparking debate over 'commoditization' of models.
GPT-5 Is Here! Free for All, Powered by the Simplest Interface
OpenAI launches GPT-5 with unified routing and fast-response modes to capture the default chat entry point. Day-one routing glitches spark user experience controversies.
Winning an IMO Gold Medal with Prompt Engineering: Tsinghua Alumni Uncover New Findings, Proving Academia Can Rival Tech Giants Without Heavy Spending
Google's IMO-level math reasoning model is now open for trial. Its performance at the Math Olympiad level has become a focal point of summer discussions.
DeepSeek's Next-Gen Tech Leaked Early; Liang Wenfeng Wins ACL2025 Best Paper Award
DeepSeek's paper and academic recognition in sparse attention, reasoning systems, and cost transparency fuel the 'China vs. Silicon Valley' narrative.
Open-Source Qwen Sweeps Three Titles in a Week, Outperforming Closed Models with SOTA Reasoning and Coding
China's major model 'industry papers' and cost-reduction practices are revealed: training and inference costs, distillation, and open-source strategies become the focus of public debate.
7B Model's 'Emotional Intelligence' Rivals GPT-4o: Tencent Breaks Through Open-Domain RL Challenge, Scores Jump Fivefold
Tencent's RLVER open-domain emotional reinforcement learning addresses 2025–2026 AI industry extension topics (batch 2 archived, complementing existing timeline).
4B Small Model Surpasses Claude 4 in Math Reasoning; 700-Step RL Training Approaches 235B Performance | HKU, ByteDance Seed & Fudan University
A 4B-parameter model's math reasoning via reinforcement learning rivals large models. This 2025–2026 AI industry extension topic (batch 2 archived) complements the main timeline.