Tops GPTs 
HackerRank has released Chakra, an AI agent that conducts live interviews, observes candidate workflows and scores not only answers but judgment and “AI fluency.” The company says Chakra ran over 500,000 interviews in beta and is now generally available to customers.
On October 6, 2026, Pinterest introduced Beauty Guides, an AI feature that converts beauty Pins into salon-ready instructions. Powered by Pinterest Intelligence, the guides translate images into stylist terminology and list time, cost range and maintenance needs.
A surge of AI assistants now works through text messaging and existing apps instead of standalone downloads. Agents such as Instinct, Caddy and Fambot connect to calendars, email and services to schedule, book and follow up, with varied pricing and beta statuses.
Bevel launched eu/jev as the first System One model hosted in the EU, aimed at organisations with GDPR and data sovereignty requirements. It runs on Hetzner GPUs in Germany, does not store prompts or answers, and uses Winnow with the same API and pricing as Jev.
Basis tested GPT-6 Astra against GPT-5.6 Sol on a complex 50-tab tax workbook and found Astra completed the task in half the time and improved internal evaluation scores by about 20%. Basis also reported better intent understanding and adaptive reasoning.
Microsoft and Hugging Face published ThinkingBox, a benchmark that judges AI agents by the terminal database state they leave, not their final sentences. Across 507 tasks run 20 times each, many agents pass once but fail repeatedly, often due to tool-handling errors.
Reassign is a 24-hour circular planner with midnight at the top and noon at the bottom. The product’s description says it offers drag-to-block scheduling, an AI command palette that can rebuild plans live, and two-way sync with Google Calendar, Outlook, Todoist, and Linear.
ServiceNow CoreAI describes AutoSynthData, a pipeline that turns a model’s failures into validated training tasks grounded in an enterprise environment. The team reports gains on EnterpriseOps Gym benchmarks after fine-tuning with synthetic samples.
Codync presents an open-source, free alternative to Grok Bot, Muse and Dots. The project says it can run Claude Code, Codex, Cursor and 40+ agents locally, with approval workflows from iPhone, Mac, Linux or a terminal, under an MIT license and end-to-end encryption.
Ai2 released AstaBrief 8B, an open-weights model designed to generate cited scientific reports faster and run locally. The team reports a 3.5× reduction in end-to-end generation time (51.1s vs 178.5s) and describes training choices and data filters intended to improve citation grounding.