GPT-5.6 Sol Pricing Cut by 50%
Pricing cut 50% lets developers run complex reasoning tasks cheaper.
Daily edition
OpenAI cuts GPT-5.6 Sol pricing while Alibaba releases Qwen3.8-27B for local AI workloads
Daily edition
69 of 69 feeds responded
221 distinct stories
19 previously issued, held back
12 worth your time
Collection and selection
221 candidate events; 38 shortlisted stories; 31 with article text available; 12 selected. Counts describe different stages; clustering, validation and refills can change the shortlist.
Active filters: none. Sort: Ranked.
All 12 stories, in ranked orderMore filters
Qwen3.8-27B runs frontier-class coding agents and reasoning locally, no cloud API required
Qwen3.8-27B enables on‑device coding and multimodal agents without cloud APIs, needing ~56 GB GPU memory at FP16.
ByteDance Seed and Tsinghua AIR Introduces CUDA Agent: A Large-Scale Agentic RL System for CUDA Kernel Generation
CUDA Agent raises pass rate to 98.8% and speeds kernels 96.8% faster than torch.compile.
Developing Nemotron 3.5 Lightning NVFP4 with QAD Using NVIDIA Model Optimizer
Nemotron 3.5 Lightning NVFP4 delivers up to 4× throughput while using only 22 GB memory after QAD compression.
Same Cluster, 33 Points More Utilization: What Changed Was the Order
A constraint‑aware allocator boosts GPU utilization by up to 33 percentage points versus FIFO scheduling.
DeepSeek Harness Open Source: Everything Is a Plugin — the Bet Is an Agent Platform, Not a Product
Open-source DeepSeek Harness lets engineers build plugin-based AI agents, accelerating custom workflow development.
Also covered byMarktechpostNous Research Ships Bot Mode for Hermes Agent, Turning Agent Profiles Into a Roster of Named Bots
Hermes Bot Mode adds named bots with independent memory and skills, ready for desktop deployment under MIT license.
Anthropic watermarks Claude's output, but critics question the tradeoffs
Anthropic’s Claude watermark embeds statistical patterns, but critics argue it may degrade text quality.
Seeing beyond BMI: Estimating cardiometabolic risk with smartphone imagery
Google’s PhotoScan predicts insulin resistance from phone images with accuracy comparable to DXA scans.
Two Humanoid Robots Play a Full 11-Point Table Tennis Match — HKU and KAI's SMASH 2.0 Autonomously Serves, Rallies, and Scores
HKU and KAI’s SMASH 2.0 robots completed a fully autonomous 11‑point table tennis game.
Distributed Layerwise Offload: Scaling Toward 200B+ DiT Models Efficiently in vLLM-Omni
vLLM‑Omni’s Distributed Layerwise Offload runs a 124 GB model on 64 GB HBM, cutting host memory by 73%.
Cursor launches Origin code hosting platform as GitHub outage exposes opening in AI coding race
Cursor’s Origin offers a self‑hosted Git alternative, launched as GitHub suffered a prolonged outage.
No stories match. Active filters.