A user asked Grok Bot to build an Apple Passwords import feature, and says it produced a modified Grok Bot build with it. One self-reported X demo, no official confirmation or verified video.

GitHub's context-aware classifier cuts median secret scanning time by 86% and raises acceptance by 11.5%. github.blog/ai-and-ml/github…

A leaked eval config names GPT-6.1 Terra as the judge model alongside GPT-6.1 Sol. OpenAI's 28-day Codex improvements reportedly end next week with a new model release, but this remains unconfirmed.

A model scored 42%, then 95% after someone fixed a grader rejecting right answers — per one X post, unverified. Audit your eval grader before trusting a low benchmark score.

An iPhone “Hey Grok” wake word uses only iOS Shortcuts and Vocal Shortcuts. Update Grok Bot, add a voice-chat Shortcut, say “Hey Grok” three times, and enable running when locked. It’s one unverified user report with a concept image, not xAI or Apple confirmation.

Reuters reports SpaceX raised $40 billion in debt for GPUs, per Musk. The post links to reut.rs/4z5fjf5, but no primary article text was supplied, so the amount and terms remain unconfirmed reporting.

A developer shipped an admin dashboard for Google’s managed agents on the Google IO platform. Builders get a central place to monitor and manage agent runs, bridging the IO demo to production tooling. Capabilities beyond that central layer aren't detailed yet.

Codex's Day 20 reset is fully propagated. Tibo Sottiaux says it's done. The 25% reset for all plans comes from a quoted Kimi post, not OpenAI's confirmation.

Nanbeige4.3-33B claims SOTA agentic performance among open ~30B models. It posts 80.9 on Terminal-Bench 2.1, 65.5 on DeepSWE and 70.4 on Toolathlon-Verified, with a 512K context window under Apache 2.0. Benchmarks are self-reported by Nanbeige. huggingface.co/Nanbeige/Nanb…

Grok Bot user says he replaced Claude and ChatGPT with a chief-of-staff AI agent managing others. That team handles content, fact-check, research and calendar tasks in one place. One Varney claim, not an xAI rollout; availability unconfirmed.

Picsart shows a Seedance 2.5 superhero parkour video, starting from a prompt on its platform. Three sampled frames show a rooftop jump, neon backflip, and sunset skyline. The exact prompt isn't visible, so treat this as a promotional demo, not verified output.

Salesforce may soon charge for Claude and other external AI tools inside its apps. Businesses may move data if fees are too high. Fees and timing are unconfirmed, based on The Information report.

Kimi K3.1 Pro finished a Three.js V8 engine challenge autonomously in 31 minutes using 3.5M tokens. DeepSeek V4.1 Flash used 67.2M tokens over 48 minutes before being stopped. Both built working simulations, but the token gap is 19x.

Claude 5.5 Sonnet built a detailed Scottish Highland steam train scene entirely from code, using zero pre-made 3D assets. The procedural generation consumed 770k tokens and took about an hour, raising questions about traditional 3D asset pipelines.

FrankenForge v0.1.0 is an open-source Rust inspector that opens Safetensors LLM checkpoints without Python. The source post calls it an early prototype with EN/RU/ZH support; no repo link is shared yet.

Microsoft's EF Core team swapped the default eval judge to GPT-6.1 Terra, then reverted it to GPT-5.6 Terra due to infrastructure issues. The evals run through GitHub Copilot (copilot-sdk), not OpenAI directly.

Jev-Mem builds agent memory 6.6x faster by using a fast scorer instead of an LLM for routing. On LoCoMo, memory construction dropped from 1,044s to 158s while answer quality improved to 0.777.

NVIDIA's open-weight Nemotron 3.5 Bolt E4B and Cosmos vision model run locally on a Jetson Orin Nano 8GB, powering a humanoid robot without cloud connectivity. Reachy demonstrates real-time vision-guided dialogue in a video credited to NVIDIA AI.

A sampled demo shows Microsoft’s 9B Decision-1 solving a Rubik’s Cube in 5.9 seconds while GPT-6 Sol reaches 6/12 moves after 22.6 seconds. It’s one account’s sampled frames, not an independent benchmark.

Hermes mobile app testing is coming very soon. A monitoring account teased it with a black-and-white anime icon, but no features, platforms, or launch date are confirmed. Android is only a hope, not a plan.