AMD Quark Accelerates Diffusion Models with MXFP4 on MI350 GPUs
AMD's Quark library enables MXFP4 quantization for FLUX.1-dev image generation, achieving 1.92× speedup over BF16 eager inference on MI350 GPUs.
Software
338 articles
The tools, frameworks, and platforms developers use to build with AI — coding assistants, agent frameworks, APIs, and developer tooling, with the updates that change how you ship.
LATEST NEWS
-
-
Amazon SageMaker AI Integrates with MLflow for Unified Experiment Tracking
Amazon SageMaker AI now streams benchmark and recommendation results to MLflow, enabling real-time tracking and reducing data silos. This integration supports multiple jobs under a single experiment.
-
Amazon Launches Multi-Turn RL Infrastructure for Amazon Nova on SageMaker HyperPod
Amazon SageMaker HyperPod now supports multi-turn reinforcement learning for complex workflows, with training costs estimated at $786–$1,180 per hour.
-
Claude Code and Fable 5 Ported 2003 PC Game to Native iOS
A Google Deepmind developer used Claude Code and Fable 5 to port Command & Conquer to iOS in a few hours, running natively on ARM64.
-
Gemma-4 31B Model Serves 1.17k Tok/s on RTX 6000 Pro
Gemma-4 31B model achieves 1.17k tokens per second throughput on RTX 6000 Pro GPU during peak load, with median TTFT of 0.7 seconds.
-
Open-source tool pxpipe cuts Claude Code and Fable 5 token costs up to 70%
The open-source tool pxpipe reduces token costs for Claude Code and Fable 5 by up to 70% by hiding text in PNGs, according to developer Steven Chong.
-
AMD Introduces AgentKernelArena for AI Coding Agent Benchmarking
AMD launches AgentKernelArena, an open-source benchmarking tool, to evaluate AI coding agents on AMD Instinct GPUs, with GEAKv3 leading performance in multiple categories.
-
Meta Launches Pocket, AI-Powered Gaming App
Meta quietly launched Pocket, an AI-powered gaming app, on June 29, 2026, with 635K lifetime installs for its predecessor Gizmo.
-
Amazon SageMaker AI Introduces Multi-Turn Reinforcement Learning
Amazon SageMaker AI now offers multi-turn reinforcement learning tools to train agents for complex tasks like content moderation and support ticket resolution, using the SOP-Bench dataset for evaluation.
-
Mistral Launches Vibe Agent for Work and Code
Mistral's Vibe agent, now available, streamlines workflows with Work and Code modes, offering free to enterprise plans starting May 28, 2026.
-
Amazon SageMaker AI Accelerates Protein Design with BoltzGen
Amazon SageMaker AI enables faster protein binder design with BoltzGen, reducing a 1,000-sample campaign from 375 hours to under 2 hours.
-
Amazon Launches Open Source Model Profiler for Amazon Bedrock
Amazon Bedrock introduces an open source tool to simplify model selection across its 100+ foundation models, with daily pricing and regional availability updates.