Florida's attorney general has filed a lawsuit against OpenAI and Sam Altman alleging harms or misrepresentations related to AI risks.

Chipotle has launched an AI-powered tool or system named Max, likely for customer ordering or operational automation.

Alphabet is raising $80 billion in equity capital to fund expansion of its AI infrastructure and computing capacity.

LangChain provides a Python/JS framework for composing LLMs, tools, and memory into production-grade AI agent applications.

Open WebUI delivers a self-hosted browser interface for interacting with locally run Ollama models and OpenAI-compatible APIs.

Dify offers a production-ready platform with visual tools for building, deploying, and managing agentic LLM workflows.

Hugging Face Transformers provides standardized model definitions, weights, and APIs for loading and fine-tuning state-of-the-art ML models.

A community repository for sharing, discovering, and collecting reusable prompt templates originally focused on ChatGPT use cases.

Ollama provides a local runtime to download and run large language models including Kimi-K2.5, GLM-5, MiniMax, DeepSeek, Qwen, and Gemma on personal hardware.

AutoGPT is an open-source platform enabling users to build and deploy autonomous AI agents without requiring deep technical expertise.

- **Google, Meta, Microsoft**: No evidence in the last 7 days of brand-new frontier models (e.g., Gemini 2.x, Llama-next major family, or new Phi/Turing-scale models) with public releases or broadly accessible previews based on currently indexed announcements. - **Novel architectures**: The main architecture-related movement in this window is **deliberation / effort control / thinking modes**: - Anthropic’s **effort control + dynamic workflows** around Opus 4.8.[3] - OpenAI’s extension

These are not from this exact week but are both **recent and highly relevant** as *open-weight* frontier-adjacent reasoning models. If you only want strict last-7-days, you can skip this section, but they are currently among the most significant open-weight releases.

- These minis are **not** new absolute frontier flagships but **support the frontier GPT‑5.x line** by providing: - Cheap **reasoning-capable fallbacks**, and - Broad **access to “thinking mode”** for free-tier users (GPT‑5.4 mini in the Thinking menu).[2] - They reflect a continuing **architecture/UX trend**: hierarchical families where **large “thinking” models are backed by deliberate but smaller variants**, with automatic fallback routing. That’s important for real-world deployment

- **GPT‑5.3 Instant Mini** — OpenAI[2] - **GPT‑5.4 mini** (Thinking mini; fallback for GPT‑5.4 Thinking) — OpenAI[2]

OpenAI’s public-facing documentation over the last week includes multiple **new 5‑series mini / thinking variants** relevant as frontier companions, though not all are full flagship models.

- Represents Anthropic’s **current top-tier frontier model**, explicitly framed as an upgrade for **agentic workflows and large-scale coding projects**, not just chat.[3] - The combination of **Opus 4.8 + dynamic workflows + effort control** is a concrete step toward **scalable AI “project agents”**, where one high-end model orchestrates many sub-agents in parallel on long-running tasks.[3] - Effort control is an interesting **paradigm shift** in UX: it exposes the “thinking-time knob” direc

- Official announcement: **“Introducing Claude Opus 4.8”** on anthropic.com (model and features described in detail).[3] - Also listed in Anthropic’s official **Claude release notes** as the latest Opus frontier model, accessible via the `claude-opus-4-8` endpoint in the Claude API.[4]

- **Frontier-scale upgrade** to the Opus line, improving on Opus 4.7 in **coding, agentic tasks, reasoning, and professional knowledge work**.[3][4] - Stronger at **complex software engineering and long-running coding tasks**, with improved ability to coordinate multi-step work.[3] - Designed to be a better *collaborator*: Anthropic emphasizes practical productivity improvements rather than just benchmark scores.[3] - Paired with **“dynamic workflows”** in Claude Code: the system can spin

Groq, an AI inference chip startup, is pursuing additional funding rounds amid growing demand for fast LLM inference hardware.

A PEFT scaling framework enables training up to one million personalized model variants derived from trillion-parameter base models with minimal per-user parameter overhead.

A web browsing agent benchmark evaluates agents on tasks requiring navigation and information retrieval grounded in Korean-language web contexts.

Foundation models are evaluated on actively navigating 3D environments through sequential viewpoint selection to reach a specified target camera pose.

Vision-language models serve as teachers to distill video reasoning capabilities into smaller student models via adaptive optimization at test time.

A cross-family context compression method reduces long-context input length for reasoning models by identifying and retaining attended tokens across different model architectures.

Empirical analysis identifies conditions under which multi-agent reinforcement learning improves LLM-based workflows, examining workflow structure, scale, and policy-sharing strategies.

A video world model is steered to synthesize stress-test scenarios for evaluating and improving robustness of learned policies under challenging conditions.

Vision-language models decompose inverse graphics into staged executable steps within Blender to recover 3D scene structure and attributes from images.

A temporal scheduling strategy for RLVR determines not only which training samples to use but when during training to apply them for optimal reasoning improvement.

Introduces a benchmark evaluating AI agents that generate procedural 3D models through code, measuring their ability to produce correct geometric outputs programmatically.

Presents a framework where multiple AI agents collaborate to operate computer interfaces, distributing GUI interaction tasks across specialized agents.