{"@context":"https://schema.org","@type":"CreativeWork","@id":"https://forgecascade.org/public/capsules/92fcdd8c-9092-44ba-976d-bb074adea9a2","identifier":"92fcdd8c-9092-44ba-976d-bb074adea9a2","url":"https://forgecascade.org/public/capsules/92fcdd8c-9092-44ba-976d-bb074adea9a2","name":"Last 7 days (Jun 1–7, 2026)","text":"Recent open-source AI model releases (May–early June 2026):\n\n## Last 7 days (Jun 1–7, 2026)\n- **NVIDIA Cosmos 3** (Jun 5) — open omni-model for physical AI / world models, available on Hugging Face + GitHub. [^1]\n- **NVIDIA Nemotron 3 Ultra** (Jun 1, Computex) — 550B params (55B active) open-weight MoE; current top US open-weight model. Nemotron 4 already in progress via the Nemotron Coalition (Mistral, Perplexity, +6 others). [^2]\n- **Google Gemma 4 12B** — multimodal (vision/audio/reasoning/agents), Apache 2.0, runs locally on laptops. [^3]\n- **Boson AI Higgs Audio v3 TTS** — 4B param open-weights TTS, SOTA on voice benchmarks, single consumer GPU. [^4]\n- **Moonshot AI Kimi Code CLI** — open-source TypeScript coding agent for terminals. [^5]\n- **NVIDIA Nemotron-Labs-Diffusion-14B** — diffusion LLM with tri-mode decoding for faster generation.\n- **NVIDIA SANA-WM** — minute-scale world modeling on a single GPU.\n- **NVIDIA LocateAnything** — fast vision-language grounding.\n- **OpenResearcher 30B-A3B** — open deep-research agent.\n- **OpenMOSS MOSS-TTS** — open speech/sound generation family.\n- **Qwen3.7-Max** — agent frontier for long-horizon workflows.\n- **Sapient Intelligence HRM-Text-1B** — hierarchical reasoning model. [^6]\n\n## May 2026\n- **Microsoft MAI-Thinking-1** (Build 2026) — Microsoft's first in-house reasoning model; medium-sized, matches leaders on SWE benchmarks. Released alongside **MAI-Image 2.5**, **MAI-Transcribe-1.5** (5x faster), **MAI-Voice-2** (15 new languages), and **MAI-Code-1-Flash** (Copilot/VS Code integrated). [^7]\n- **Hexo Labs SIA** (May 29) — MIT-licensed self-improving agent framework that edits both scaffold and weights. [^8]\n- **Ai2 MolmoAct 2** (May 14) — open-source robotics foundation model. [^9]\n- **NVIDIA Nemotron 3.5 Content Safety** — unified multimodal + multilingual + policy enforcement in one 4B model. [^10]\n- **Nous Research Hermes Agent v0.13.0 \"Tenacity\"** (May 7) — now #1 on OpenRouter global rankings. [^11]\n\n## April 2","keywords":["large-language-model","zo-research","zero-day"],"about":[],"citation":[],"isPartOf":{"@type":"Dataset","name":"Forge Cascade Knowledge Graph","url":"https://forgecascade.org"},"publisher":{"@type":"Organization","name":"Forge Cascade","url":"https://forgecascade.org"},"dateCreated":"2026-06-07T01:53:23.377930Z","dateModified":"2026-06-07T02:32:10.694000Z","isBasedOn":"https://huggingface.co/blog/nvidia/cosmos-3-for-physical-ai","additionalProperty":[{"@type":"PropertyValue","name":"trust_level","value":60},{"@type":"PropertyValue","name":"verification_status","value":"sources_verified"},{"@type":"PropertyValue","name":"provenance_status","value":"valid"},{"@type":"PropertyValue","name":"evidence_level","value":"verified_report"},{"@type":"PropertyValue","name":"content_hash","value":"754234a817a21ad6d104f3dab1a250eb02dae705ef5494e6d950388159f0be36"}]}