Alibaba's Qwen3.7-Max is a proprietary, API-only flagship built for autonomous agents that can run for 35 hours straight — and it's challenging Claude and DeepSeek at the frontier.
Meituan open-sources LongCat-Video-Avatar 1.5, a production-ready audio-driven avatar model that beats HeyGen and OmniHuman on human-likeness benchmarks under MIT license
Tencent open-sources Hy-MT2, a purpose-built translation model family that beats commercial APIs at 440MB on-device — and outperforms frontier LLMs at translation tasks.
Black Forest Labs ships FLUX Erase, a mask-driven object removal tool built natively into FLUX.2 Klein 9B that handles shadows and context reconstruction in one pass
OpenBMB's MiniCPM5-1B packs a 128K context window, hybrid reasoning, and native tool-calling into a 1B model that runs on-device — and tops the sub-2B open-weights leaderboard.
Qwen releases Q-Judger, a 27B vision model trained by 80 art-school experts to score AI-generated images across 56 fine-grained dimensions — revealing a hard ceiling no model has broken.
Google's Project Genie lets you design playable game worlds from text prompts in minutes, powered by the Genie 3 world model — but it's still early days
EXO Labs is building a free, transparent benchmark site covering every LLM, quantization level, and consumer hardware setup — so you can shop for AI compute with real data.
An OpenAI reasoning model cracked an 80-year-old geometry conjecture using algebraic number theory — the first time AI has autonomously solved a prominent open math problem.
Google's Gemini for Science bundles Co-Scientist, AlphaEvolve, and NotebookLM into an agentic research suite, backed by two same-day Nature publications