Halogen 0.17.2 hits ~45 tps with Qwen 3.8 Flash Next on Strix Halo

AWS details pay-per-inference for AI agents, where purchases are small and frequent — sometimes a fraction of a cent each — and happen inside the agent's loop with no person available.

NVIDIA's KGMON team placed second in the KDD Cup 2026 Data Agents competition, building its system around a smaller, more verifiable agent harness rather than a larger model. Teams had to work with a small, fixed LLM, making the harness the main optimization surface.

OpenAI told investors its annualized revenue is approaching $50 billion, per the FT, roughly $20 billion less than the previously reported $70 billion. The higher figure came from investors' attempts to compare directly with Anthropic's annualized revenue, which counts cloud-partner sales that OpenAI does not.

A community developer built Baikal LoopSR x2, a tiny anime/illustration upscaler that applies the recurrent shared-block concept from Looped-DiT to direct RGB restoration. Released for ComfyUI with weights available.