Image created with gemini-3.1-flash-image-preview with claude-sonnet-4-5. Image prompt: Using the provided reference image, preserve the exact weathered crate construction with dark reddish-brown paint, horizontal slats, iron hardware, and three-panel hand-painted black stencil layout, but replace the text with ‘OPEN SOURCE’ in the same loose brushstroke style. The crate sits on a stone doorstep in early spring light with its wooden lid pried partially open, revealing glimpses of circuit boards and hand-labeled components inside, a crowbar leaning against the side, soft focus background, photorealistic 1950s material world.
A huge release from OpenClaw: > ClawHub-first installs + native skills search/install/update > New plugin SDK, Matrix plugin, Claude/Codex/Cursor bundle support > GPT-5.4 default, GPT-5.4-mini/nano, MiniMax M2.7 + per-agent reasoning > /btw side questions > unified core image
https://x.com/TheTuringPost/status/2036360384882577555
Cohere Transcribe: state-of-the-art speech recognition https://cohere.com/blog/transcribe
Speaking of Voxtral | Mistral AI https://mistral.ai/news/voxtral-tts
Ai2 just released MolmoPoint GUI on Hugging Face A specialized VLM for GUI automation that points using grounding-tokens instead of coordinates, reaching 61.1 on ScreenSpotPro.
https://x.com/HuggingPapers/status/2036101402477404284
Black Forest Labs, Cursor, LangChain, Mistral AI, Perplexity, Reflection AI, Sarvam AI and Thinking Machines Lab – what unites these companies? Just recently NVIDIA announced the Nemotron Coalition, gathering all of them to develop the Nemotron family of models. → The idea is
https://x.com/TheTuringPost/status/2035320446124695922
hf-mount Attach any Storage Bucket, model or dataset from @huggingface as a local filesystem This is a game changer, as it allows you to attach remote storage that is 100x bigger than your local machine’s disk. This is also perfect for Agentic storage!! Read-write for Storage
https://x.com/julien_c/status/2036436553082286342
🎉 Congrats to @MistralAI on launching Voxtral 4B TTS — enterprise-grade TTS built for production voice agents. Day-0 support in vLLM Omni. 🌍 9 languages with natural prosody and emotional range 🎙️ 20 preset voices with easy adaptation to new ones ⚡ Ultra-low latency
https://x.com/vllm_project/status/2037193518519902408
Fleet now has shareable skills. Capture your team’s domain knowledge once, attach it to any agent, and share it across your workspace. Create skills from a prompt or previous chat, write them manually, or use a template. Read more: https://x.com/LangChain/status/2036858148850671903
Give coding agent a sharable workspace with persistent storage. With hf-mount + Codex/Claude Code you can give any agent a cloud-backed project directory. Mount a @huggingface bucket, point your agent at it, done. MOUNT_DIR=/tmp/agent hf-mount start bucket user/agent
https://x.com/ben_burtenshaw/status/2036827952588234783
Cohear 👂 Cohere’s first audio model. Apache 2.0. #1 on the Open ASR leaderboard. Multilingual transcription across 14 languages.
https://x.com/aidangomez/status/2037172942803701838
🎉 Congrats to @Cohere on releasing Cohere Transcribe, a 2B speech recognition model (Apache 2.0, 14 languages). Day-0 support in vLLM. Cohere contributed encoder-decoder serving optimizations to vLLM: variable-length encoder batching and packed attention for the decoder. Up to
https://x.com/vllm_project/status/2037197243111895066
Introducing: Cohere Transcribe – a new state-of-the-art in open source speech recognition.
https://x.com/cohere/status/2037159129345614174
New SoTa transcription model from @cohere! – #1 on accuracy on the Open ASR Leaderboard. – Open Source (Apache 2.0) – 14 Languages (English, French, Arabic, German, Italian, Spanish, Portuguese, Greek, Dutch, Polish, Chinese, Japanese, Korean, Vietnamese).
https://x.com/JayAlammar/status/2037172878165053951
🔊Introducing Voxtral TTS: our new frontier open-weight model for natural, expressive, and ultra-fast text-to-speech 🎭Realistic, emotionally expressive speech. 🌍Supports 9 languages and accurately captures diverse dialects. ⚡Very low latency for time-to-first-audio. 🔄Easily
https://x.com/MistralAI/status/2037183026539483288
Mistral AI released Voxtral TTS, a 3-billion-parameter text-to-speech model with open weights that the company says outperformed ElevenLabs Flash v2.5 in human preference tests roughly 63% of the time on standard voices and nearly 70% on voice customization. The model runs on
https://x.com/kimmonismus/status/2037149838023024753
Our first speech model, Voxtral TTS, is out. It delivers SOTA performance while significantly reducing cost compared to existing solutions, and it operates with very low latency. It uses a new architecture that combines auto-regressive generation of semantic speech tokens with
https://x.com/GuillaumeLample/status/2037274172607594609
Cursor admits its new coding model was built on top of Moonshot AI’s Kimi | TechCrunch https://techcrunch.com/2026/03/22/cursor-admits-its-new-coding-model-was-built-on-top-of-moonshot-ais-kimi/
Kimi.ai on X: “Zhilin at GTC: Introducing Attention Residuals Learning selective memory, rather than mechanically accumulating everything, is the beauty of attention. Many of you have probably read Attention Is All You Need, the 2017 Transformer paper that brought “human-like” attention into https://t.co/1coOW90s0n” / X
https://x.com/Kimi_Moonshot/status/2037010118957817988
Ksenia_TuringPost on X: “Deep transformers used to accumulate layer history. Now they are starting to retrieve from it. → @Kimi_Moonshot proposed Attention Residuals (AttnRes), driving this shift. They turn the residual stream into an attention problem. Why do we need it? Depth in Transformers mostly https://t.co/L4pMwyiRY2” / X
https://x.com/TheTuringPost/status/2037107923109953788
was messing with the OpenAI base URL in Cursor and caught this accounts/anysphere/models/kimi-k2p5-rl-0317-s515-fast so composer 2 is just Kimi K2.5 with RL at least rename the model ID
https://x.com/fynnso/status/2034706304875602030?s=20
Closed Source vs Open Source AI: A Cage Fight Few People Understand https://davefriedman.substack.com/p/closed-source-vs-open-source-ai-a
Did you feel that vibe shift anon? Open Source is in the air.
https://x.com/NousResearch/status/2036122143398961659
God I love Open Source. Integrated browser into T3Code Gonna add a terminal next.
https://x.com/LLMJunky/status/2035856842224497049
Our models are now available on @OpenRouter 🧠🦾 Our brand new Edge, designed for the constraints of the physical world: ⚡️️Sub-second latency 📉 Optimized compute 🏗️ Production-ready reasoning Deploy frontier intelligence closer to your data. Try it now:
https://x.com/RekaAILabs/status/2037186645246530025
This is why we made T3 Code open source btw. Love this!
https://x.com/theo/status/2036216034949312851
Today, we’re launching the Resend CLI • 53 commands • Fully open source $ curl -fsSL https://resend.com/install.sh | bash
https://x.com/zenorocha/status/2032459310341800314?s=20
Is the Future of AI Local? | Tom Bedor’s Blog https://tombedor.dev/open-source-models/
Local AI is free, fast & secure! So today we’re introducing hf-mount: attach any storage bucket, model or dataset from @huggingface as a local filesystem. This is a game changer, as it allows you to attach remote storage that is 100x bigger than your local machine’s disk. This
https://x.com/ClementDelangue/status/2036452081750409383
Now available on Hugging Face: hf-mount 🧑🚀 The team really cooked, still wrapping my head everything possible but you can do things like: – mount a 5TB dataset as a local folder and query only the parts you need with DuckDB (✅ works) – browse any model repo with ls/cat like
https://x.com/victormustar/status/2036476453370380416
WebGPU is INSANE! 🤯 Here’s a 24B parameter model running locally in a web browser, at a blazing ~50 tokens/second on my M4 Max. ⚡️ It’s the largest model we’ve ever run with Transformers.js… and we’re not stopping here. Big announcement soon.
https://x.com/xenovacom/status/2036908326462665211
On the Direction of RLVR Updates for LLM Reasoning | Notion https://qwen-pilot.notion.site/rlvr-direction
Qwen3.5-REAP-262B running with 200K context inside Parchi, controlling multiple browser tabs reviewing Sitegeist features against Parchi. For free 180GB VRAM TTFT is pretty slow Generation 36 tokens/s Prefill 1400 tokens/s Automate anything w Parchi https://x.com/0xSero/status/2036204079056081043
This is looking like a good prediction. Alibaba’s Qwen and Xiaomi both seem to be steering away from open weights in the 2 weeks since this post.
https://x.com/emollick/status/2034983183653986337
Leave a Reply