Swiftlet: Run an 80B Qwen Locally on a Mac or iPhone
Swiftlet runs the real Qwen3-Next-80B on a Mac in 4.3 GB of RAM and a 35B model on an iPhone, using expert streaming to keep only active experts in memory.
Articles
In-depth analysis and deep dives into the AI tools shaping creative work.
Google Cloud API Gateway Now Routes to Any LLM
Google shipped model routing for Cloud API Gateway on August 4, 2026: one OpenAI-compatible endpoint that forwards each request to Gemini, Claude, or an OpenAI GPT model.
Velokey: One API for Kling, Veo, Sora, and Seedance
Velokey is a new pay-as-you-go API gateway that unifies Kling, Veo, Sora, Seedance, plus image and LLM models behind one OpenAI-compatible endpoint.
NVIDIA VoiceChat-11B: Open Full-Duplex Voice Model
NVIDIA released open weights for VoiceChat-11B, the first open full-duplex voice model that calls tools mid-conversation. Here is how it works and how to self-host it.
Voice Coding Tools 2026: On-Device vs Cloud Compared
If you code under an NDA or on a private repo, an on-device voice coding tool is the only safe way to talk to your AI agent. SKI, Spokenly, and VoiceMode keep every word local. Here is how the leading voice coding tools compare.
FFmpeg 9.0 Lei: GPU Acceleration for Video Creators
FFmpeg 9.0 Lei shipped August 3, 2026, the most GPU-focused release in years: animated WebP decode, Vulkan 360 filtering, ProRes RAW on Apple, and AMD AMF HDR filters.
Cloudflare Cuts Kimi and GLM Costs 30% on Workers AI
Cloudflare shipped three inference optimizations that make Kimi K2.6 and GLM 5.2 roughly 30% cheaper to run on Workers AI, and its benchmark tables show the answers do not change.
Qwen3.8-Max: Alibaba's Open-Weight Coding Flagship
Alibaba announced Qwen3.8-Max on August 3, 2026: a 2.4 trillion-parameter coding flagship with a 1 million-token context and a promised open-weights release.
EU AI Act Article 50: How to Label AI Content Now
As of August 2, 2026, the EU AI Act Article 50 transparency rules are enforceable. Here is what creators must disclose and how to label AI-generated content.
LoRA Dataset Studio: One-Tab Local LoRA Trainer
LoRA Dataset Studio is a free, self-hosted web app that runs the entire LoRA training lifecycle inside a single browser tab, from one reference photo to a trained, ranked model.
Grok Imagine 1.5 Adds 1080p and Voice References
xAI expanded Grok Imagine Video 1.5 with native 1080p, text-to-video, and up to seven locked image and voice references, live August 1 on web, iOS, Android, and the API.
audio.cpp 0.5: Run a Local Audio Studio, No Python
audio.cpp 0.5 turns a pure C++ engine into a full local audio studio: expressive DramaBox TTS, Confucius4 cross-lingual voice transfer, music, and AMD ROCm support, all with no Python.
Ideogram P-Image vs GPT-Image vs FLUX: Cost Compared
Ideogram P-Image starts at $0.003 per image with Ideogram signature text rendering. Here is how it compares to GPT-Image, FLUX, Ideogram 4.0, and Qwen on price, speed, and in-image text.
QM: YC Open-Sources Its Multi-Agent Team Harness
Y Combinator open-sourced QM, an MIT-licensed multiplayer agent harness that gives every teammate an isolated workspace and drives Claude Code, Codex, OpenCode, and Pi from one core.
DeepSeek V4 Flash: Open-Weight Coding Agent Model
DeepSeek shipped the official production build of V4 Flash on July 31, 2026: an open-weight, MIT-licensed 284B-parameter MoE model with a 1M token context, native OpenAI Responses API and Codex support, and API output at $0.28 per million tokens.
MiniMax H3 Brings 2K AI Video With Native Sound
MiniMax released H3 on July 31, 2026: native 2K video with built-in stereo sound, all-modal input, and a price it says runs at a third of rival models.
Figma Make Gets a Properties Panel and Annotations
Figma Make just added a properties panel and annotations, letting designers edit AI-generated apps visually instead of describing every tweak in a chat box.
NVIDIA NOOA: Build AI Agents as Python Objects
NVIDIA open-sourced NOOA, a model-agnostic framework that turns an AI agent into an ordinary Python object. Here is how it works, the benchmarks, and how to build your first agent.
How to Design a Logo and Brand Identity With AI
A step-by-step workflow to design a logo and starter brand identity with AI in about an hour, using a text-capable generator, a vectorizer, and free type and color tools.
Perplexity Projects: A Multiplayer AI Workspace
Perplexity Projects turns Computer into a shared, multiplayer workspace for people and AI agents, with a persistent file system and Brain memory that carries context between tasks.
Cursor for iPad: Review and Merge AI Agents From a Tablet
Cursor shipped a dedicated iPad app on July 29, 2026, letting you run, review, and merge AI coding agents with split screen, pinned sidebar chats, and Apple Pencil markup, no laptop required.
Google Lyria 3.5: Better Vocals in Flow Music
Google DeepMind released Lyria 3.5 in Flow Music on July 29, 2026, upgrading vocals, lyrics, musicality, and tempo control. Here is what changed and how it compares to Suno and Lyria 3 Pro.
Replit Design: AI Creative Suite for Builders
Replit launched Replit Design on July 29, 2026, an AI creative suite that takes an idea to a live app with Mobbin references, brand systems, and no handoff. How it works and how it compares.
Run Gemma 4 26B on a Mac in 2GB RAM: TurboFieldfare
TurboFieldfare is an open-source Swift and Metal engine that runs Google Gemma 4 26B in about 2GB of RAM on any M-series Mac by streaming experts from SSD.