Artificial Analysis adds live GPU rental prices to its agent benchmark, turning raw throughput numbers into a per-dollar scoreboard for Blackwell, Hopper and MI355X systems.
A new arXiv paper argues the OCR step in document research agents is both unnecessary and worse than feeding page images straight to a vision-language model.
A 2.8GB distilled reasoning model gets stripped of safety refusals through automated abliteration, running locally on consumer hardware with near-zero refusal rate.
Tencent's Hunyuan Research Agent moves beyond one-off problem solving into autonomously building reusable scientific models, workflows, and software tools.
Pika Labs open-sources LoRA adapters that distill MiniMax H3 from 50 steps down to 6, generating 15-second video with synced audio in roughly 7 seconds on an H100.
A Rust rewrite of SearXNG ships as a single binary with 24x faster startup, 9x less memory, and drop-in config compatibility for AI grounding workloads.
A new interface technique from Penn and Microsoft Research turns citation numbers into unfoldable evidence trails, driving measurably deeper engagement with AI answer sources.