Google unveils SL2T, converts sign language to text, rolling out with Pixel 11
Instead of typing, deaf users can sign, and Gboard and Live Transcribe turn it into text
70
AI
AI models, services and robotics
Relationship-management app Mesh expands to Android, adds AI contact search
The personal CRM app acquired by Automattic launches on Android, testing Beeper integration and an AI contact-discovery feature
20
DeepSeek-V4-Pro moves to general availability, API pricing revealed
$0.435 input, $0.87 output per million tokens with 1M-token context — cache hits cut costs sharply
30
Grok 4.6 scores 61 on intelligence index, matches GPT-5.6 at the same price
Score jumps 5 points a month after launch, takes the runner-up spot to Claude Opus 5 in agentic tasks
40
Cohere unveils 2.4B-parameter vision model 'North Micro Vision'
Ultra-compact VLM specialized in document and chart understanding, released free on Hugging Face under Apache 2.0
40
Liquid AI unveils screen-reading model that runs in 3GB
LFM2.5-VL-3B hits 20 tokens/sec on Galaxy S26 Ultra, 228 tokens/sec on M5 Max
10
Google Pixel 11 price rises $100, Gemini gets more embedded
At Made by Google 2026, Google unveils Pixel 11, Watch 5, and 'Pixel Tag' to rival Apple AirTag
50
Google Trains Gemini's Clinical Skills Through Simulated Residency
ResidencyRL applies reinforcement learning to Gemini 3.5 Flash using 81 conditions and 50,000 simulated consultations
40
Solo Operator Runs Company With Nine Grok Bots
AI For Success operator unveils a Grok Bot squad dividing up briefings, scheduling, and receivables management
280
Gemini Enterprise Testing Chat-Task Toggle
Google is testing a chat/task split and a marketing campaign agent in Gemini Enterprise
20
Hidden reasoning tokens from Claude, GPT recovered — benchmark memorization suspected
A paper reportedly exposed 100% of the hidden chain-of-thought from closed APIs, reigniting the benchmark reliability debate
40
ChatGPT credit gifting feature spotted in Help Center
Eligible customers buy credits and send a one-time link; recipients redeem it on their personal account
10
Unitree passes 18,000 humanoid robots in cumulative production
The count covers bipedal robots only — a sign that hardware manufacturing is scaling up alongside AI brain progress
20
Higgsfield unveils 'Layers,' a tool that auto-splits poster designs into editable layers
Headlines, badges, and people that used to take hours to separate by hand can now be pulled apart in seconds
80
Microsoft's new coding model falls short of DeepSeek on both price and performance
MAI Code 1.1 Flash improves on its predecessor but still trails the open-source DeepSeek V4 Flash
70
Sakana AI teases robot fish with "Mecha Fish" preview image
A newsletter promo image shows a robot fish under development alongside the phrase "autonomous shoaling"
80
Tencent Hunyuan Publishes Five-Level Framework for Verifying AI Agent "Self-Evolution"
Survey paper breaks down into five levels the evidence needed to judge whether an agent that modifies itself has actually improved
50
Zuckerberg admits "Llama 4 wasn't it" as Meta open-sources Muse
In interview, Zuckerberg admits misstep, releases Muse Glimmer weights and teases Spark 1.2
40
Databricks Launches Unity AI Gateway to GA, Aims to Curb "Token Waste"
Unity AI Gateway moves to general availability, using real-time budget policies to control model selection
80
Factory Builds AI Dev Environment Where Code Never Leaves the Machine, on DGX Spark
By running NVIDIA Nemotron 3.5 Lightning on local hardware, the company automates software development without exposing source code
20
MiniMax Code updates browser automation and goal mode
v3.0.54 adds form-filling and file-upload automation plus long-task tracking
40
Ant Group's Ling 3.0 Tiny scores 25 on intelligence index with 1.3B active parameters
Its accuracy was the lowest in its class, but its hallucination rate was also the lowest at 30%. The tradeoff: 50,000 tokens per answer.
70
Vercel v0 redesigns sidebar with real-time task status in chat
AI coding tool v0 adds project-based chat grouping and git branch previews to its sidebar
10
LTX-2.5, open video generation model runs on a single consumer RTX GPU
An open-weight world model that produces a 10-second clip in 6.8 seconds, optimized for NVIDIA GPUs
180