Guides
How to compare AI models
Start with your actual task, not a leaderboard. Compare answer quality, total cost, speed, language support and data handling under the same conditions.
In short
Oossa's checklist: use the same examples and record the model version, date, settings and errors. Separate company claims from independent tests. This guide is a method, not a product ranking.
Oossa · Published by Oossa:
Produced and translated with AI assistance. Check the original sources below.
Models →
Today
Models
Transformers 5.18.0 adds streaming speaker diarization model
Hugging Face’s new release includes Nemotron 3 Diarization, a streaming model that can label up to eight speakers in live audio.
Models
Runway unveils Praxis‑1, a generalist robot policy model
Runway Research announced Praxis‑1, an open‑weight robot action model trained on video, with early tests by Noble Machines, Standard Bots and Ultra.
Models
Blackfuel says it has $250 million in contracted revenue
The company says it is building a global AI computing network for running models, but has not shared details about its customers or infrastructure.
Models
llama.cpp fixes CPU handling for BF16 matrix inputs
Release b11292 adds CPU support for a BF16 matrix input used by depthwise convolution and tightens Vulkan’s support checks.
Models
Google reports 2.4× faster sparse video attention on TPUs
Google says a tile-aligned attention kernel cut latency on one TPU v6e chip. The reported speedup covers the attention kernel, not the full video-generation pipeline.
Models
Hugging Face launches Open TTS Leaderboard for faster model evaluation
The new leaderboard uses objective metrics to rank open‑source speech models in hours instead of weeks.
