AI Intel

DeepSeek-V4-Pro API is offering a 75% discount until May 2026, and allows access to a 1M context window by setting the model to `deepseek-v4-pro[1m]`.

Alibaba's high-performance 125B MoE model, Qwen3.8-Flash, can now run locally with just 75GB RAM, reportedly outperforming Claude-Opus-4.6 (Max).

Weave Router 2.0 has been launched on Product Hunt, presenting itself as a subscription-aware router designed for coding agents.

Xiaomi Mimo 2.6 now includes a live post-training dashboard, allowing users to monitor model training progress in real-time.

Thread, an AI-powered journal, has launched on Product Hunt, designed to help users connect their thoughts and ideas into a more comprehensive narrat…

Toki Coordination, launched on Product Hunt, serves as a personal AI assistant focused on streamlining meeting scheduling and follow-up processes.

Research shows a 4B model can be trained to produce query plans 81% faster than Postgres, indicating significant advancements in AI-driven database o…

A Mozilla report suggests China's open-weight AI models are only four months behind leading US offerings, providing a cheaper alternative despite lag…

llama.cpp has released update b11009, which includes a fix for split state and granularity issues in fused QKV for Gemma4 and Qwen35 models.

Appwrite 2.0 has been launched on Product Hunt, positioning itself as an open-source cloud platform specifically designed for agents and developers.