AI MODEL MONITOR
AI Model Release Monitor — Daily Digest (2026-10-07)
2026-10-07
2 new model release(s) found.
OTHER RELEASES
EmbeddingGemma 2 (Google DeepMind)
Google Blog (official announcement)
A 740M multimodal embedding model that maps text, code, images, audio, and video into a shared vector space for cross-modal search and local retrieval. Its modular encoders support text-only deployments and offline/on-device use.
Hardware: ~1.6GB VRAM at full precision (FP16/BF16), ~0.8GB at INT8, ~0.4GB at 4-bit (GGUF/AWQ) quantization. Runs comfortably on a single consumer GPU (8-12GB) or even CPU/laptop RAM at Q4. Google reports about 567MB active RAM for its optimized quantized full-model on-device deployment.
Mistral Large 4 (le Chonk) (Mistral AI)
Mistral AI official announcement
Mistral’s natively multimodal 1T-parameter MoE (49B active) targets coding, agentic workflows, multimodal understanding, and enterprise work. Public preview is currently API-only; Mistral says downloadable weights are planned by the end of October.
Hardware: ~2.1TB VRAM at full precision (FP16/BF16), ~1100GB at INT8, ~550GB at 4-bit (GGUF/AWQ) quantization. Effectively datacenter-only – multiple 80GB+ GPUs even at 4-bit quantization; not practical on consumer hardware. (MoE: only ~49B active per token, so inference speed is much faster than the total size implies, but full weights still need to be loaded.) No public weights yet; this is a rough estimate for the announced 1T-parameter size if weights arrive as planned.
Full browsable, searchable model page
Source: https://drive.google.com/file/d/1NpL8_FQv2YM5fM2x8N9A5w8jgfi3xrjQ/view?usp=sharing
Source: https://drive.google.com/file/d/1NpL8_FQv2YM5fM2x8N9A5w8jgfi3xrjQ/view?usp=sharing
RAW1 AI — BUILDS AUTOMATED INTELLIGENCE SYSTEMS.