AI Model Updates
Gemma 4: Our most intelligent open models to date, purpose-built for advanced reasoning and agentic workflows.
<p>Article URL: <a href="https://github.com/SharpAI/SwiftLM">https://github.com/SharpAI/SwiftLM</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47597181">https://news.ycombinator.com/item?id=47597181</a></p>
<p>Points: 1</p>
<p># Comments: 2</p>
<p>Article URL: <a href="https://twitter.com/OpenRouter/status/2038701599175196715">https://twitter.com/OpenRouter/status/2038701599175196715</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47578983">https://news.ycombinator.com/item?id=47578983</a></p>
<p>Points: 5</p>
<p># Comments: 0</p>
<p>Article URL: <a href="https://aistudio.google.com/app/new_music?model=lyria-3-pro-preview">https://aistudio.google.com/app/new_music?model=lyria-3-pro-preview</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47566347">https://news.ycombinator.com/item?id=47566347</a></p>
<p>Points: 2</p>
<p># Comments: 0</p>
<p>I was personally invited by the Qwen team to speak at Qwen Meetup Korea, and got to present locally here in Korea yesterday — pretty honored to have been reached out to directly.<p>The talk was about how I got function calling to work reliably on deeply recursive union types — the stuff the industry generally says doesn't work. With qwen3-coder-next, first-try success rate was 6.75%. And the entire Qwen 3.5 model family was hitting 0% on union types due to a consistent double-stringify b...
<p>Article URL: <a href="https://medium.com/google-cloud/1-million-tokens-per-second-qwen-3-5-27b-on-gke-with-b200-gpus-161da5c1b592">https://medium.com/google-cloud/1-million-tokens-per-second-qwen-3-5-27b-on-gke-with-b200-gpus-161da5c1b592</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47542691">https://news.ycombinator.com/item?id=47542691</a></p>
<p>Points: 3</p>
<p># Comments: 0</p>
Our latest voice model has improved precision and lower latency to make voice interactions more fluid, natural and precise.
Google DeepMind researches AI's harmful manipulation risks across areas like finance and health, leading to new safety measures.
Introducing Lyria 3 Pro, which unlocks longer tracks with structural awareness. We’re also bringing Lyria to more Google products and surfaces.
<p>I think it's much better than before, and might even replace perplexity.</p>
<hr>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47513343">https://news.ycombinator.com/item?id=47513343</a></p>
<p>Points: 7</p>
<p># Comments: 3</p>
<p>Article URL: <a href="https://gist.github.com/kibotu/a009f00414b7c10fb1c74e603d7838c0">https://gist.github.com/kibotu/a009f00414b7c10fb1c74e603d7838c0</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47499324">https://news.ycombinator.com/item?id=47499324</a></p>
<p>Points: 2</p>
<p># Comments: 0</p>
<p>Article URL: <a href="https://twitter.com/danveloper/status/2034353876753592372">https://twitter.com/danveloper/status/2034353876753592372</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47498414">https://news.ycombinator.com/item?id=47498414</a></p>
<p>Points: 2</p>
<p># Comments: 1</p>
<p>Article URL: <a href="https://blog.google/innovation-and-ai/technology/developers-tools/full-stack-vibe-coding-google-ai-studio/">https://blog.google/innovation-and-ai/technology/developers-tools/full-stack-vibe-coding-google-ai-studio/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47450049">https://news.ycombinator.com/item?id=47450049</a></p>
<p>Points: 2</p>
<p># Comments: 3</p>
<p>Article URL: <a href="https://www.testingcatalog.com/minimax-launches-m2-7-model-on-minimax-agent-and-apis/">https://www.testingcatalog.com/minimax-launches-m2-7-model-on-minimax-agent-and-apis/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47446452">https://news.ycombinator.com/item?id=47446452</a></p>
<p>Points: 2</p>
<p># Comments: 0</p>
<p>Article URL: <a href="https://twitter.com/GoogleAIStudio/status/2034654985850659149">https://twitter.com/GoogleAIStudio/status/2034654985850659149</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47442421">https://news.ycombinator.com/item?id=47442421</a></p>
<p>Points: 6</p>
<p># Comments: 0</p>
<p>Article URL: <a href="https://github.com/huanglizhuo/QwenASR">https://github.com/huanglizhuo/QwenASR</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47438212">https://news.ycombinator.com/item?id=47438212</a></p>
<p>Points: 1</p>
<p># Comments: 0</p>
<p>Article URL: <a href="https://simonwillison.net/2026/Mar/18/llm-in-a-flash/">https://simonwillison.net/2026/Mar/18/llm-in-a-flash/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47438166">https://news.ycombinator.com/item?id=47438166</a></p>
<p>Points: 3</p>
<p># Comments: 0</p>
<p>Article URL: <a href="https://old.reddit.com/r/LocalLLaMA/comments/1rwvn6h/minimaxm27_announced/">https://old.reddit.com/r/LocalLLaMA/comments/1rwvn6h/minimaxm27_announced/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47434629">https://news.ycombinator.com/item?id=47434629</a></p>
<p>Points: 1</p>
<p># Comments: 0</p>
<p>I kept trying to use their M2.5 model and now they released M2.7, but they are TERRIBLE.<p>See this comparison I made:<p>https://aibenchy.com/compare/minimax-minimax-m2-7-medium/minimax-minimax-m2-5-medium/z-ai-glm-5-medium/google-gemini-3-1-flash-lite-preview-medium/<p>Not only that, but M2.5 is #1 on OpenRouter, which is crazy: https://openrouter.ai/rankings<p>I think the only reason why it is #1 is because it is a scam. In the comparison you can see it had over 200k reasoning tokens, wher...
<p>Article URL: <a href="https://openrouter.ai/minimax/minimax-m2.7">https://openrouter.ai/minimax/minimax-m2.7</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47425518">https://news.ycombinator.com/item?id=47425518</a></p>
<p>Points: 7</p>
<p># Comments: 1</p>
We’re introducing a framework to measure progress toward AGI, and launching a Kaggle hackathon to build the relevant evaluations.
<p>Article URL: <a href="https://fararoni.dev/publicacion/caso-estudio-qwen">https://fararoni.dev/publicacion/caso-estudio-qwen</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47401192">https://news.ycombinator.com/item?id=47401192</a></p>
<p>Points: 1</p>
<p># Comments: 1</p>
<p>Article URL: <a href="https://discuss.ai.google.dev/t/google-ai-pro-subscription-antigravity-quota-not-working-as-advertised-10-day-lockout-instead-of-5-hour-reset/118505">https://discuss.ai.google.dev/t/google-ai-pro-subscription-antigravity-quota-not-working-as-advertised-10-day-lockout-instead-of-5-hour-reset/118505</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47374098">https://news.ycombinator.com/item?id=47374098</a></p>
<p>Points: 1</p>
<p># Comments: 0</p>
<p>I was chatting with MiniMax M2.5 in OpenRouter and suddenly he mysteriously repeated on "I'm Claude, an AI assistant created by Anthropic - not a "language" ", heh wut?</p>
<hr>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47372273">https://news.ycombinator.com/item?id=47372273</a></p>
<p>Points: 12</p>
<p># Comments: 11</p>
<p>Article URL: <a href="https://twitter.com/antigravity/status/2031835833716625883">https://twitter.com/antigravity/status/2031835833716625883</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47348613">https://news.ycombinator.com/item?id=47348613</a></p>
<p>Points: 3</p>
<p># Comments: 0</p>
<p>Article URL: <a href="https://ianlpaterson.com/blog/llm-benchmark-2026-38-actual-tasks-15-models-for-2-29/">https://ianlpaterson.com/blog/llm-benchmark-2026-38-actual-tasks-15-models-for-2-29/</a></p>
<p>Comments URL: <a href="https://news.ycombinator.com/item?id=47324730">https://news.ycombinator.com/item?id=47324730</a></p>
<p>Points: 3</p>
<p># Comments: 1</p>
<p>I wanted a way to summarize YouTube videos without paying for a SaaS or leaking my viewing history to someone. TubeTrim is a Python-based tool that runs LLMs locally to process transcripts. No API keys, no subscriptions, no tracking.<p>It uses the transformers library with a device-aware backend: it will prioritize CUDA, then MPS (for Mac users), and finally fallback to CPU. I've found that Qwen 2.5-1.5B provides a good balance between speed and summary quality for this specific task.<p>How ...