The open sparse MoE model for agentic coding Open Source • Artificial Intelligence • Development 1 126 Wispr Flow: Dictation That Works Everywhere Stop typing. Start speaking. 4x faster. Productivity • Artificial Intelligence • Audio Tanay Kothari Hey Product Hunt 🎉 I’m Tanay, co-founder & CEO of Wispr Flow, a Mac dictation app that lets you speak naturally, and writes in your style, in every application — with auto-edits, command mode, and over 100 languages. ⭐ The founding story Ever since I watched the first Ironman movie in 2008 at age 10, I've wanted to build Jarvis. It drove me to pull my first all-nighter and start teaching myself how to code. 16 years later, I think we’ve found a way to make a voice experience delightful for all-day use. And there's no one else I'd build this with than my college roommate and closest friend @sahaj_garg2 Our vision is to make voice interfaces both useful (so you trust them) and ubiquitous (so you can use them everywhere). This is how we can move from screen-first technologies to voice-first technologies and build a future where we aren't stuck looking at our phones all day long. ⭐ Using Flow is super simple: 1. Download Flow for Mac 2. Press and hold [Fn] to start speaking in any app 3. Release [Fn] to enter text ⭐ What users love Flow for - 🛠️ Developers using Cursor / Claude / ChatGPT: Speak with AI assistants faster than typing. - ✉️ Professionals breezing through their inbox: Flow accurately captures names and formats your emails and Slack messages. - 🧑🎓 Students chatting with AI and finishing assignments even faster: We have a special student discount. - 📄 Product Managers drafting PRDs and sharing thoughts: Flow turns your rambles into clear ideas. - 👶 Parents with busy lives: Time is precious. Every second you save writing is an extra second you have for family. - 🤖 Tech-lovers who want to use voice with every AI tool ⭐ Here’s all you’ll get with Flow 1.0 — and we’re just getting started. 1. ⚡ Blazing fast dictation: Powered by Flow’s ultra-fast inference engine 2. 🎨 Tone match: You speak differently than you write. Flow learns your writing style across every application 3. 🔧 Auto-edits when you change your mind: “Hey lets meet at 5pm, actually lets do 6pm” → “Hey, lets meet at 6pm” 4. 😎 Command Mode for selected text: Say commands like “Flow, make this crisper and more assertive” without copy-pasting into other tools. 5. 🧩 Native integrations: Select text anywhere and just say: “Ask perplexity, what does this mean?” 6. 😶 Whispering mode: Use Flow around others by quietly whispering to your computer. 7. 🔒 Private by design: Your recordings locally on your computer by default. Only you have access to it. You can allow Flow to use your data to improve our models (disabled by default). Learn more: wispr.ai/data-usage For technical users, most voice dictation tools focus on technical metrics like "word error rate." At Flow, we prioritize what truly matters to users: zero-edit messages. With Flow, you rarely need to return to your keyboard for edits. Our new approach has made Flow the first consumer voice dictation platform that makes people enjoy using voice more than their keyboards. ⭐ A final note Our dream is to create a world where interacting with technology feels as natural as interacting with people. I'd love your help to make this a reality. Try out Flow and share your feedback — we're eager to make it even more magical with your input. PS: A huge shoutout to our thousands of beta users who've showered us with love and feedback over the last few months. We wouldn't be here without you.
Qwen3 is the large language model series developed by Qwen team, Alibaba Cloud. - QwenLM/Qwen3
Qwen3-30B-A3B provides both thinking and non-thinking modes, enabling developers to build applications requiring complex logical reasoning, mathematics, coding, and general-purpose chat. The model supports 256K-token context (extendable to 1M tokens) and 100+ languages, allowing flexible deployment for diverse reasoning tasks.
As an open-source model with publicly available weights, Qwen3-30B-A3B can be deployed on self-hosted infrastructure using frameworks like Transformers, llama.cpp, Ollama, or vLLM. This eliminates dependency on proprietary APIs and allows full control over data and deployment.
Qwen3-30B-A3B's dual thinking/non-thinking modes allow developers to benchmark different reasoning strategies within a single model. The technical report and evaluation results enable informed decisions about model selection for specific tasks like mathematics, code generation, or general chat.
The 30B-A3B variant balances model capability with computational efficiency, making it suitable for integration into production applications. Developers can use it with quantization techniques (GPTQ, AWQ) to reduce memory footprint and inference costs while maintaining strong performance on reasoning and coding tasks.