In a 10-person exoskeleton pilot, walking patterns fitted to each user scored no better than an averaged pattern. How long people had been in the device mattered more.
Red Hat's 4-bit NVFP4 build of Gemma 4 26B-A4B cuts memory needs by about 75%. It keeps 97 to 99% of the original model's accuracy on most benchmarks, but agentic tool use and hard math take bigger hits.
Tencent has released the weights for Hy4 flash, a 220B-parameter multimodal MoE model with 12B active parameters and a 1M-token context, plus FP8 and base versions.
Pooling more training corpora can make speech deepfake detectors worse. IDFE adds a corpus classifier with gradient reversal so the model can't lean on dataset quirks, and it cuts pooled error by up to 20%.
GPT-6.2 Astra tops Vals AI's new multi-turn coding benchmark, which checks whether agents can keep extending a working app through up to ten requests without breaking existing features.
The ZazenCodes Season 3 repo is trending because of its Claude Code "software factory" templates. They turn labeled GitHub issues into reviewed pull requests, and a human still does the final merge.
Microsoft's first speech-to-speech model undercuts GPT Realtime 2.1 on price and claims higher accuracy, with OpenAI Realtime API compatibility for easy migration.
Perplexity has opened up the model routing layer it uses in production. One API key now reaches 45+ open and proprietary models, with automatic provider failover, one bill, and an optional router that picks a model for each request.
Microsoft AI drops seven models at Ignite that undercut GPT and Claude on price while matching performance on everyday coding, image editing, voice and security tasks.