The last seven days brought a concentrated wave of AI releases. This roundup covers major foundation and specialist models officially announced between August 27 and September 3, 2026. It excludes community fine-tunes, rumors, and product-only updates.
The releases
- Gemini 3.8 Flash — Google's new fast workhorse targets coding, multi-step reasoning, and long-running agent workflows at a lower cost than larger frontier models.
- Gemini 3.8 Flash Cyber — A security-focused variant for vulnerability discovery and automated patching, available through Google's restricted Fairwind Program.
- Claude Fable 5.1 — Anthropic's generally available flagship for coding, research, and long-running knowledge work, with cheaper cache reads than Fable 5.
- Claude Mythos 5.1 — The same core model with fewer cyber and biology restrictions, limited to vetted research and defense organizations.
- GLM-5.3-Flash — Z.ai's natively multimodal mixture-of-experts model uses 320B total parameters but activates only 18B per token for efficient agentic work.
- Quasar 438B — Multiverse Computing's bilingual English-Spanish reasoning model is designed for enterprise agents and coding.
- Tencent Hy4 preview — An open-weight 770B-parameter model with 49B active parameters and a context window above one million tokens, focused on coding and productivity.
- Gemini Omni 1.1 Flash — Google's updated multimodal video model adds stronger creative control and production-ready generation from text, image, audio, and video inputs.
- Mistral OCR 4.1 — Now generally available, this document model adds paragraph bounding boxes, structural labels, and block-level confidence scores.
What matters
The common direction is clear: models are becoming more specialized, more agentic, and more efficient. Cybersecurity now has dedicated restricted-access models, while open and sparse architectures are pushing longer context windows without activating every parameter.
Bottom line
This was not one winner-takes-all launch week. It was a shift toward purpose-built intelligence: faster coding models, controlled cyber capability, stronger multimodal generation, and practical document automation.
