Foundation Model

Kimi Linear: 6.3× Faster 1M-Token Decoding, 75% Less KV Cache

Kimi Linear: 6× faster, 75% leaner attention—rewriting Transformer limits.

Oct 31 •

MiniMax Open-Sources M2, Free Limited-Time

10B-active MiniMax-M2 tops open-source charts—code, agents, free API & weights

Oct 27 •

GLM-4.6 Beats Claude 4 in Coding, MIT Open-Source

GLM-4.6: Beat Claude 4 in coding tests. 27% faster, 30% less tokens. MIT open-source.

Oct 10 •

ByteDance's First Inference Model Leads in 7 Rankings

ByteDance's Seed-OSS-36B: an open-source 36B-parameter model with 512K context, tops 7 benchmarks. Apache-2.0 license. Free for research & commercial…

Aug 21 •

Mistral Exposed for Distilling DeepSeek and Faking Results

Mistral accused of distilling DeepSeek's model & faking benchmarks. Whistleblower exposes alleged misconduct. Open-source AI giant faces controversy.

Aug 14 •

GLM-4.5 Technical Report Released: Full Training Details Unveiled

Zhipu AI releases GLM-4.5 technical report: ARC model details, MoE architecture, slime RL framework & benchmark dominance.

Aug 11 •

World's First Large-Scale Visual Memory Model Released

Memories.ai launches world's 1st Large Visual Memory Model! Outperforms Google & OpenAI in tests. Get a free 1st-month trial now!

Jul 26 •

Le Chat Unveils Voxtral: World's Top Open-Source Voice Model

Mistral AI's Voxtral: World's best open-source voice model. Le Chat now rivals ChatGPT with voice, research & image editing features. Europe's AI rising…

Jul 19 •

Google Releases T5Gemma, Reigniting the Architecture War!

Google releases T5Gemma, reviving encoder-decoder architecture with superior performance over decoder-only models in AI reasoning tasks.

Jul 14 •

First Global Medical Code Generation Training Platform Launches

MedAgentGym: World's first medical code generation training platform. Open-source Med-Copilot-7B matches GPT-4o performance on medical programming…

Jul 13 •

Kimi Open-Sources Trillion-Parameter K2 Model, Pressuring OpenAI

Kimi K2 open-source trillion-parameter model challenges GPT-4 & Claude with superior coding abilities at $2/million tokens. New SOTA benchmarks.

Jul 12 •

Baidu Open-Sources ERNIE 4.5 Series, Outperforms DeepSeek-V3

Baidu open-sources 10 ERNIE 4.5 models with MoE architecture. Features 424B parameters, multimodal capabilities, and outperforms competitors on…

Jun 30 •

#nojs-banner { position: fixed; bottom: 0; left: 0; padding: 16px 16px 16px 32px; width: 100%; box-sizing: border-box; background: red; color: white; font-family: -apple-system, "Segoe UI", Roboto, Helvetica, Arial, sans-serif, "Apple Color Emoji", "Segoe UI Emoji", "Segoe UI Symbol"; font-size: 13px; line-height: 13px; } #nojs-banner a { color: inherit; text-decoration: underline; } This site requires JavaScript to run correctly. Please turn on JavaScript or unblock scripts