AI Models & APIs
-
Isomorphic Labs Drug Design Engine (IsoDDE) doubles AlphaFold 3’s accuracy on unseen protein-ligand structures
Isomorphic Labs, the DeepMind spinout, finally answered the question everyone had after AlphaFold 3: what comes next. The answer is IsoDDE, and it hit the HN front page today. What it actually is Not a chatbot, not an API you can sign up for. IsoDDE is a model system — one unified engine with four… Continue reading
-
Satya Nadella calls Claude Fable 5 “editorially controlled” in front of his own Copilot engineers
Nadella, to the engineers building Microsoft’s Copilot: “If you use Fable, when it refuses for any random thing, it just is like, when was the last time you had a creation tool that was so editorially controlled?” His verdict on Anthropic’s refusal policy: “It doesn’t make sense.” What Fable 5 is, and what set this… Continue reading
-
Moonshot AI ships Kimi K3: 2.5T parameters, 1M context, zero benchmarks published
Kimi K3 went live the night of July 16. No model card, no license, no pricing, no benchmark table. Just a flagship model appearing on kimi.com and a timeline losing it. What’s confirmable: a Mixture-of-Experts model around 2.5 trillion parameters, 1M-token context, text plus images and audio. A general-purpose chat and coding model, aimed straight… Continue reading
-
Inkling (Thinking Machines Lab): Mira Murati’s first model ships as a 975B open-weight MoE you can download
Mira Murati left OpenAI, raised a fortune, and stayed quiet for a year. On July 15 the silence broke: Thinking Machines Lab dropped Inkling, its first public product — and the whole thing is on HuggingFace under Apache 2.0. What it actually is Inkling is a foundation model, not an app. 975B total parameters, 41B… Continue reading
-
ByteDance Seedream 5.0 Pro outputs 10+ transparent PNG layers, no manual cutouts
Every AI image generator hands you a flat JPEG. You want to move the logo? Too bad, it’s baked into the pixels. ByteDance’s new flagship image model, launched July 8, kills that problem. What Seedream 5.0 Pro actually does It’s a text-to-image model with one killer trick: layered output. One render splits into a background… Continue reading
-
PrismML’s Bonsai 27B squeezes a 27B multimodal model into 3.9GB — and runs it on an iPhone
On-device LLMs have been stuck at 3B–8B for years. PrismML, a Caltech spinout backed by Khosla, Google and Samsung, just shipped Bonsai 27B: a Qwen3.6 27B multimodal model crushed down to 5.9GB (ternary) or 3.9GB (1-bit). That 3.9GB number is not an accident — it’s roughly the app memory budget iOS gives you. HN put… Continue reading
-
Apple SpeechAnalyzer API benchmark: 55% faster than Whisper, and more accurate in English
The first real third-party benchmark of Apple’s SpeechAnalyzer API just hit 369 points on HackerNews, and the result is blunt: for English transcription on current iPhones and Macs, “just use Whisper” is no longer the default answer. The numbers A 34-minute 4K video: SpeechAnalyzer transcribed it in 45 seconds. Whisper Large-v3 Turbo took 101 seconds.… Continue reading
-
Gemini API Managed Agents 大更新：后台任务 + 远程 MCP，免费层开放 — Google’s hosted agents go free
Google shipped four upgrades to Managed Agents in the Gemini API on July 7, and the headline isn’t any feature — it’s the price. The whole thing now runs on the free tier. Quick recap: Managed Agents is an API where Google runs the agent for you inside a cloud sandbox — reasoning, code execution,… Continue reading
-
Mesh LLM pools every GPU in your house into one local AI cluster — in an 18MB binary
n0, the team behind P2P networking library iroh, shipped Mesh LLM on July 11. It grabbed 147 points on the Hacker News front page in a day. The pitch: your gaming PC, your MacBook, that old workstation — stitched into one inference cluster, no cloud involved. How the mesh actually works This is local-first inference… Continue reading
-
Mistral Robostral Navigate hits 76.6% on R2R-CE with one RGB camera
Mistral shipped Robostral Navigate on July 8: an 8B open-weight robotics navigation model. Give a robot a plain-language instruction — “go to the kitchen and grab the red cup” — and it navigates unfamiliar indoor spaces with a single RGB camera. No LiDAR, no depth sensor, no pre-built map. Why 8B and one camera matter… Continue reading
Top AI Product 