-
Apple AFM 3 runs a 20B model on your iPhone — and hands it to developers
At WWDC 2026 Apple shipped AFM 3, its third-generation Foundation Models: five models split across on-device and cloud. The headline is AFM 3 Core Advanced, a 20-billion-parameter sparse model that actually runs on an iPhone. The flash-memory trick A 20B model won’t fit in phone RAM, so Apple doesn’t try. The full model lives in… Continue reading
-
ByteDance Doubao Seed 2.1 Pro / Turbo undercuts Claude Opus 4.6 on price — and claims to beat it on code
ByteDance shipped two new flagship LLMs on June 24: Doubao Seed 2.1 Pro and a half-price Turbo twin. These aren’t chatbots in the consumer sense — they’re API models tuned for coding, long-horizon agent tasks, and multimodal work. The pitch is blunt: frontier capability, Chinese pricing. What it actually does Pro leads Terminal Bench 2.1,… Continue reading
-
OpenAI GPT-5.6 government-managed access list: ~20 partners, approved one by one
This isn’t a product. It’s a precedent. OpenAI shipped GPT-5.6 — the Sol, Terra, and Luna models — but the U.S. government decided who gets to touch it first. About 20 companies made the preview, each cleared by Washington on a customer-by-customer basis. What’s actually new here A frontier lab releasing its best model through… Continue reading
-
OpenAI GPT-5.6 (Sol / Terra / Luna): a three-tier lineup only 20 orgs can touch
OpenAI shipped GPT-5.6 on June 26 — not one model but three. Sol is the flagship, OpenAI’s most capable model yet. Terra is the mid-tier workhorse. Luna is the cheap, fast one. Same day, GPT-4.5 quietly vanished from ChatGPT. The catch: you can’t have it. OpenAI handed the models to the U.S. government first and… Continue reading
-
RubyLLM gives Ruby one interface for every major LLM provider
Ruby developers have spent years stitching together one-off gems for each AI provider. RubyLLM collapses all of that into a single framework: one chat object, one config block, and the same code whether you’re calling OpenAI, Anthropic, Gemini, Bedrock, DeepSeek, Mistral, Ollama, or a local model. Switching from gpt-4o to claude-sonnet-4-5 is a one-liner. ##… Continue reading
-
Qualcomm buys Modular for $3.9 billion to drain Nvidia’s CUDA moat
Qualcomm is buying Modular in an all-stock deal worth about $3.9 billion — and the target isn’t a chip, it’s software. Modular makes the Mojo programming language and the MAX inference platform, the closest thing the industry has to an open answer to Nvidia’s CUDA lock-in. That’s what Qualcomm actually paid for. ## What Modular… Continue reading
-
AI labs are hiring philosophers, not just engineers, to decide how models behave
The hottest non-engineering hire at a frontier AI lab right now might be a philosopher. Anthropic, Google DeepMind, and OpenAI are bringing philosophers in-house to decide how their models should behave — a trend drawing fresh attention as AI capability outruns anyone’s settled view of AI ethics. ## Who’s hiring whom DeepMind is bringing on… Continue reading
-
HackerRank hiring-agent scores resumes by digging into your GitHub
Resume screening is the part of hiring everyone hates and nobody does well. hiring-agent, an open-source tool from HackerRank (interviewstreet) trending on GitHub, automates it — and notably, it doesn’t just read the resume, it goes and checks your code. ## How it works The agent converts a resume PDF to markdown with PyMuPDF, then… Continue reading
-
DESIGN.md gives coding agents a structured spec for your visual identity
Coding agents are great at writing components and terrible at remembering your brand — colors, spacing, and tone drift every session. DESIGN.md, an open format from Google Labs, tries to fix that by giving agents one persistent, structured description of a design system. ## What it is A DESIGN.md file is a plain-text source of… Continue reading
-
Orca is an agent development environment for running a fleet of parallel coding agents
One coding agent is useful; a dozen running at once is chaos to manage. Orca, from Stably, is an ADE — an agent development environment — built to run a whole fleet of coding agents in parallel, on desktop and mobile. ## What it does Orca is agent-agnostic: if it runs in a terminal, it… Continue reading
Top AI Product
Every day, hundreds of new AI tools launch across Product Hunt, Hacker News, and GitHub. We dig through the noise so you don't have to — surfacing only the ones worth your attention with honest, no-fluff reviews. Explore our latest picks, deep dives, and curated collections to find your next favorite AI tool.
Top AI Product 