Web research: on 1,266 BrowseComp tasks, Computer reaches 66.7% accuracy vs 50.2% for Pi and 43.9% for Hermes on their respective search providers. Portable uses the least wall time and the fewest tokens. Inference and private documents stay local; only search touches the web.

Web agent economics are set by inference volume and each step is one inference call. A simple task (extract a field, fill a form, navigate a site) is 10 to 20 calls. A multi-site workflow runs into the hundreds. From @RaiseSummit last month: @abhshkdz, @yutori_ai co-founder/CEO, on how they ship Navigator at 2-5x cheaper and faster than closed models on web tasks: -> ~80% prefix cache hit rate -> Speculative decoding on smaller footprint models -> Post-training Qwen models with SFT and RL They run on Together, where customizable endpoints and auto-scaling let them A/B test new model variants in minutes.

Grok web just got a new Library page You can now keep track of everything you’ve created with Grok in one place: • Media • Apps • Files • Uploads You can search through everything, filter by type and jump straight back into anything you’ve made You now don't have to dig through old chats trying to find that one image, app or file