Perplexity Lily: 1.35x Faster Local AI Than MLX on Mac
Perplexity open-sources Lily, a Rust and Metal inference engine that beats MLX by 1.35x on decode for Qwen3.6-35B-A3B on Apple Silicon. Code is ...
Fuzzball 4.2: AI Agents Can Now Run HPC Jobs via MCP
CIQ shipped Fuzzball 4.2 with a native MCP server. Any AI agent that speaks Model ...
OpenCode v2 Beta: Rewrite Fixes Memory, Model Lock-In
OpenCode v2 beta rewrites the AI coding agent from scratch: Bun to Node, Tauri to ...
Grok Bot for Enterprise: What Developers Must Know Now
Grok Bot is not a coding assistant. It’s a persistent digital worker that signs into ...
Nvidia Acquires Hugging Face: What Developers Must Know
Nvidia's $12.93B Hugging Face acquisition closes in 2027. Here's what the deal means for open-source ...
vLLM Speculative Decoding: 2.87x Faster on AMD GPUs
Speculative decoding in vLLM now hits 2.87x throughput on AMD MI300X GPUs. Learn which method to pick, where AMD is cost-viable, and where ...
Claude Code Weekly Limits Drop 17% on September 14
Anthropic announced a 25% Claude Code limit increase, but current users get a 17% cut. ...
CrowdStrike Falcon Guardian: AI Agent Runtime Security Arrives
CrowdStrike's Falcon Guardian brings runtime enforcement to AI agents — blocking shadow agents, tracking MCP ...
DeepSeek Harness (dsh): 200K Stars, No Model Lock-In, One Command to Run
DeepSeek Harness (dsh) hit 200K GitHub stars in weeks. MIT-licensed, model-agnostic, and runs with one ...
Qwen 3.8 27B on Cerebras: 1,500 Tokens Per Second
Cerebras now serves Qwen 3.8 27B at 1,500 tokens per second. Here is the pricing ...