Created by Cheng Niu
Open‑source and flagship AI model releases, benchmarks, safety notes across LLMs, vision, speech, multimodal
Explore the latest content tracked by AI Model Release Tracker
Elephant Alpha, a new stealth 100B instant model, heralds the Model Tsunami with more releases incoming. Its relatively small size means it won't rival giants, but signals accelerating AI proliferation.
Excitement builds as Moonshot AI announces Kimi K2.6 is on the way, with the KIMI Code Beta Program session ended.
Now Kimi K2.6-code-preview is...
Claude Mythos Preview achieves SOTA cyber evals, uncovering a 27-year undiscovered OpenBSD vuln (now patched) and thousands of high/critical...
WildDet3D sets new SOTA in open-world monocular 3D detection, hitting 22.6/24.8 AP3D on WildDet3D-Bench with text/box prompts and +20.7 AP from depth...
Key launch details:
Mythos Preview leads cyber benchmarks, marking a leap over prior models:
Open-source demo unpacks MolmoAct-7B for depth-aware reasoning, visual trajectory tracing, and robotic action prediction from multi-view images.
-...
Key highlights from Anthropic's launch:
YuelDesign revolutionizes drug design with diffusion models that craft molecules for flexible, moving proteins – unlike rigid models – capturing...
RefineAnything, a multimodal diffusion model, enables region-specific image refinement—fixing local details like distorted text while perfectly...
Anthropic's frontier model Claude Mythos Preview dominates cybersecurity, finding zero-days in OSes/browsers like a 27-year-old OpenBSD bug and...
Anthropic's frontier breakthrough—Claude Mythos Preview crushes benchmarks at 93.9% on SWE-bench Verified and 83.1% on CyberGym, dwarfing prior models...
CT-1 vision-language-camera models innovatively transfer spatial reasoning knowledge to enable camera-controllable video generation.
Grok tanks hardest among 8 mainstream LLMs predicting Premier League results: lost all $133k once, failed twice, avg pot zero.
Inference-time scaling enhances language model reasoning by extending chain-of-thought (CoT), yet existing approaches rely on a single policy.
AI2's MolmoWeb drops as an 8B open-source web agent using screenshots to operate browsers—like humans do—changing the game for agents.
Amid dense...
New domain-specific vision framework CLEAR leverages cognitive LLM for adaptive restoration in complex maritime scenarios.
Major multimodal boost for Gemini: generates interactive 3D models and simulations directly in chat for physics/chemistry concepts.
