Fetching from the wire…
01
02
03
04
05
06
07
08
Models2026-09-03 · source-backed
Muse Voice Transcribe, released September 1, is a single real-time model doing streaming ASR, speaker diarization and endpointing at $3.00 per 1,000 audio minutes, roughly 80% below Google Cloud Speech-to-Text's standard $0.96/hour. It handles 20+ speakers natively with no post-processing and reports 17.5% average diarization error across AMI-IHM, AMI-SDM and VoxConverse. On Artificial Analysis's AA-WER Streaming it scored 3.1% word error rate, ahead of Cartesia Ink-2 at 3.4%, ElevenLabs Scribe v2 Realtime at 3.6% and Gemini 3.5 Transcribe Live at 4%.
Each link below shares sources, entities, or timing with this story.
Launched August 8 as the new Quality Mode at grok.com/imagine and in the Grok mobile apps, pitching precision editing, crisp text rendering, and improved factuality, with API access promised but not shipped (The Decoder). On the August 7 Arena leaderboards the faster "low" var...
As of August 5, anyone pinned to the alias gets a different model. 82.9% on Artificial Analysis' speech-to-speech benchmark versus 75.7% for 1.0, against GPT-Realtime-2.1 at 79.1% and Gemini 3.1 Flash at 69.5%, with first-audio response down from 1.25s to roughly 0.70s and med...
Launched August 26, supporting more than 85 languages with accents and dialects, word-level timestamps, diarization for up to three speakers (more is experimental), custom vocabulary and jargon recognition, and automatic removal of filler words and self-corrections. Google cit...
Luu's September 1 post drew 852 points and over 1,000 comments, walking predictions from February 2024 through November 2025 after removing unfalsifiable and tautological ones. Misses include "AI has peaked" (Feb 2024), Meta "dying" (Nov 2024) against revenue going $135B to $2...
Designed with Broadcom, built by TSMC, going into manufacturing after a six-week test run found no major issues (Reuters). Iris supplements rather than replaces Meta's Nvidia/AMD GPUs and supports a buildout targeting 7GW by end-2026 and 14GW in 2027 against up to $145B in 202...
arXiv 2608.05108 skips the RL-trained attacker models that dominate red teaming and generalize poorly, instead accumulating a strategy library across a sequence of (dataset, target) pairs that transfers to unseen targets with no retraining. AgentDojo: 86.7% ASR against Gemini-...
MindPattern daily
One email a day at 7 AM. Sources and a take on every story. Unsubscribe anytime.