Results for “Open models”

Why it matters

Alibaba has open-sourced Page Agent, a client-side TypeScript library that executes natural language browser commands directly in the live page. It reads the DOM as text via DOM dehydration, bypassing external drivers like Playwright.

Why it matters

Anthropic published an evaluation of Zhipu AI's open-weight model GLM-5.3, showing it builds end-to-end cyber exploits at a rate similar to Claude Mythos Preview. Anthropic warns the model was released without meaningful safeguards and that attackers can bypass them 64% to 100% of the time.

Why it matters

Poolside launched Laguna S 2.1, an open-weight model specialized for coding agents. Available for free local hosting via vLLM and Ollama, hosted API prices start at $0.09 per 1M input tokens and $0.18 per 1M output tokens.

Why it matters

Thinking Machines Lab released Inkling Small, an open-weight multimodal mixture-of-experts model with 12B active parameters out of 276B total. It achieves a 40.2 Artificial Analysis intelligence score and offers self-hosting options alongside cheap API pricing of $0.50 input and $1.20 output per million tokens.

Why it matters

RunOnMine is an open-source, local-first security gateway for Model Context Protocol (MCP) agents running on macOS, Linux, and Windows. It restricts AI tool access to explicitly designated directories, requires local approval for risky commands, and enforces isolated browser profiles.

Why it matters

Open Science is an open-source, local-first research workbench built with Tauri and Model Context Protocol. It allows developers to run autonomous agent loops that generate reproducible notebooks, reports, and artifacts locally.

Why it matters

Apple introduced updated Mac Studio and Mac mini desktops featuring M5 Ultra and M6 chips. Offering up to 512GB of unified memory and 1.2TB/s bandwidth, the hardware targets developers running large open-weights models locally via MLX.

Why it matters

Alibaba released Qwen3.8-27B under an unrevokable Apache 2.0 open source license. Featuring native vision and a 262,000 token context window, it fits within 17 GB of VRAM for consumer hardware execution.

Why it matters

Research reveals that standard chat templates act as an on/off switch for model disclaimers like 'I am an AI' across 8 open instruct models up to 9B parameters. Removing an isolated activation steering direction turns this disclaimer voice down, while a random direction of the same size has little effect.

Why it matters

Aleph Alpha released Kolibri, an Apache 2.0 open-weight model with 78.1B total parameters and only 3.46B active per token. Featuring a custom UniBPE tokenizer, it cuts token counts for compound languages like German while matching frontier context scales.

Why it matters

Salesforce and NVIDIA unveiled Koa, an open-weight reasoning model post-trained on NVIDIA Nemotron for enterprise workflows. Designed for the Agentforce platform, it targets sales, customer service, and marketing tasks using synthetic data to lower token costs without leaking customer records.

Why it matters

Telnyx launched hosting for Moonshot AI's flagship Kimi K3 model, featuring 2.8 trillion parameters, a 1M-token context window, and native vision. The model supports function calling and OpenAI-compatible API endpoints.