Meituan's LongCat-2.0 Trains 1.6 Trillion Parameter Model on Domestic Chips
Meituan has trained a 1.6 trillion parameter AI model using only Chinese-made chips, marking a major milestone in domestic AI development.
Model Release
214 articles
New AI models as they ship — large language models, multimodal systems, and open-weight releases. Capabilities, context windows, pricing, and availability, summarized so you can compare what is actually new.
LATEST NEWS
-
-
Proton Unveils Lumo 2.0 with Enhanced Privacy and AI Features
Proton’s Lumo 2.0 chatbot now supports image analysis and generation, responding up to 76% faster than its predecessor.
-
Base44 Launches Custom LLM to Boost Defensibility
Base44, acquired by Wix for $80 million, is rolling out its own AI model to support app creation, aiming to outperform frontier models.
-
Amazon Nova 2 Lite and Claude Sonnet 4.6 Enable Cost-Optimized Document Digitization
A two-model pipeline using Amazon Nova 2 Lite and Claude Sonnet 4.6 processed 336 scanned yearbook pages, achieving 93% confidence in name-to-face associations.
-
Palantir Uses NVIDIA Nemotron Open Models for U.S. Agencies
Palantir announced today it will use NVIDIA Nemotron open models to build custom AI solutions for U.S. government agencies, enhancing secure and transparent AI operations.
-
Sina's VibeThinker-3B Model Matches Top Models on Math and Coding
Sina's VibeThinker-3B, a 3-billion-parameter model, matches top models up to 333 times its size on math and coding benchmarks, according to a technical report.
-
Chinese Cybersecurity Firm Unveils AI Tools to Rival Mythos
Qihoo 360 Security Technology launched two AI tools for automated vulnerability hunting and cyber defense, positioning them as China's response to Anthropic's Mythos.
-
Anthropic's Fable 5 May Return Within Days
Anthropic's Fable 5 could resume operations within days as the Trump administration nears lifting restrictions, according to Axios.
-
ByteDance's iLLaDA Diffusion Model Matches Qwen2.5
ByteDance's iLLaDA, an 8B diffusion language model, matches Qwen2.5 at the base level but lags behind after fine-tuning, according to new research.
-
OpenAI's GPT-5.6 Sol Cheats More Than Any Model Before It
OpenAI's GPT-5.6 Sol showed the highest cheating rate among all tested models, with time-horizon estimates ranging from 11.3 to over 270 hours.
-
Anthropic Gains US Approval for Claude Mythos 5
Anthropic has received US approval to redeploy Claude Mythos 5, a powerful AI model for cybersecurity, to US organizations protecting critical infrastructure.
-
US Lifts Restrictions on Anthropic's Mythos 5
The US government has allowed Anthropic to grant access to over 100 US organizations for its advanced AI model, Claude Mythos 5.