AI Intel

Ollama released v0.34.4, which improves structured outputs for 'thinking models' to be faster and more reliable, and fixes several issues including i…

Agent Arena has launched Pareto frontier analysis to compare model performance against median task cost for real-world agentic tasks, highlighting to…

Agent Arena's latest evaluation shows Kimi K3 (Max) to be highly cost-effective for agentic tasks with a median cost of $0.62, significantly lower th…

Google Gemini Code Assist rolls out updates, including the General Availability of Gemini 2.5 Pro and Gemini 2.5 Flash models, a new agent mode previ…

DeepSeek-V4-Pro API is offering a 75% discount and now supports 1M context, allowing users to select the deepseek-v4-pro[1m] model in Claude Code.

Anthropic is partnering with Accenture for independent evaluations of frontier AI, with both companies expecting to invest over $1 billion in buildin…

Anthropic CEO Dario Amodei has published an essay advocating for the AI industry to slow down frontier AI development, outlining a three-part plan, w…

OpenAI has shared a new framework for tracking, investigating, and disclosing instances of model misalignment, setting criteria and timelines for pub…

Mistral AI has announced a partnership with Mozilla to enhance privacy, control, and choice for users browsing online with AI.

A Reddit user reported that the Qwen-3.8-27B model performs well enough that they have stopped using paid APIs.