AI Intel

A recent update or development in `llama.cpp` has reportedly led to a 42x speed improvement for prompt lookup drafting.

OpenAI Codex has released its `0.159.0-alpha.9` version, indicating ongoing development in their code generation models.

A discussion or research paper explores how the use of specific chat templates, such as prompts starting with "As a Language Model," can significantl…

Ollama `v0.40.0` introduces a significant change where models with supported architectures will automatically run on the MLX runtime when deployed on…

A Reddit user is looking for recommendations on the optimal `qwen3.8 27b` model for systems with 16GB VRAM, aiming to balance speed, context window,…

OpenAI reportedly expressed concerns about the public perception and "optics" of information that might surface or be discussed on platforms like Hac…

Humalike x GTA RP has launched, featuring advanced AI NPCs capable of independent speech, memory, and actions, enhancing immersive role-playing exper…

DeepSeek-V4-Pro API is available at a 75% discount until May 2026 and now integrates with Claude Code to unlock 1M context, alongside OpenCode update…

Arena.ai's analysis of top agent models in Agent Arena revealed that Kimi K3 (Max) offers superior cost-efficiency, with a median task cost of $0.62,…

Arena.ai has launched a Pareto frontier analysis in Agent Arena, allowing users to evaluate agent model performance on real-world tasks against their…