Image created with OpenAI GPT-Image-1. Image prompt: over-the-top 1990s pro-wrestling promo poster, unlocked-padlock ramp featuring “Open-Source Outlaw” tossing free-license scrolls into the crowd; golden padlock pyro, grainy print texture, vivid neon titles
🚀 Introducing Qwen3-MT – our most powerful translation model yet! Trained on trillions of multilingual tokens, it supports 92+ languages—covering 95%+ of the world’s population. 🌍✨ 🔑 Why Qwen3-MT? ✅ Top-tier translation quality ✅ Customizable: terminology control, domain https://x.com/Alibaba_Qwen/status/1948406830688018471
🚀 We’re excited to introduce Qwen3-235B-A22B-Thinking-2507 — our most advanced reasoning model yet! Over the past 3 months, we’ve significantly scaled and enhanced the thinking capability of Qwen3, achieving: ✅ Improved performance in logical reasoning, math, science & coding https://x.com/Alibaba_Qwen/status/1948688466386280706
Less than two weeks Kimi K2’s release, @Alibaba_Qwen’s new Qwen3-Coder surpasses it with half the size and double the context window. Despite a significant initial lead, open source models are catching up to closed source and seem to be reaching escape velocity. https://x.com/cline/status/1948072664075223319
Qwen COOKED – beats Kimi K2 and competitive to Claude Opus 4 at 25% total parameters 🤯 https://x.com/reach_vb/status/1947357343101960424
Qwen3-235B-A22B scored 41% on ARC-AGI-1 without thinking! That’s the same level as Gemini 2.5 Pro, Sonnet 4 or o3-low with thinking. But it might be trained on it, if not, then it’s insane”” / X https://x.com/scaling01/status/1947351789222711455
RT @itsPaulAi: Wait so Alibaba Qwen has just released ANOTHER model?? Qwen3-Coder is simply one of the best coding model we’ve ever seen.…”” / X https://x.com/ClementDelangue/status/1947775783067603188
RT @lmstudio: Qwen/Qwen3-Coder with tool calling is supported in LM Studio 0.3.20, out now. 480B parameters, 35B active. Requires about 25…”” / X https://x.com/huybery/status/1948327670493970534
The new Qwen3 update takes back the benchmark crown from Kimi 2. Some highlights of how Qwen3 235B-A22B differs from Kimi 2: – 4.25x smaller overall but has more layers (transformer blocks); 235B vs 1 trillion – 1.5x fewer active parameters (22B vs. 32B) – much fewer experts in https://x.com/rasbt/status/1947393814496190712
The updated Qwen3-235B-A22B is now the best non-reasoning models period. It beats Kimi-K2, Claude-4 Opus and DeepSeek V3 on multiple benchmarks like GPQA, AIME, ARC-AGI, LiveCodeBench or BFCLv3, just to name a few. https://x.com/scaling01/status/1947350866840748521
So to recap: – Yesterday, frontier closed model equivalent reasoning model from Qwen, – This morning, frontier closed model equivalent reasoning vision capabilities from stepfun – sometime today(?) a frontier video model from wan? All open source What is America doing?”” / X https://x.com/Teknium1/status/1948744914876920039
It’s time for the American AI community to wake up, drop the “”open is not safe”” bullshit, and return to its roots: open science and open-source AI, powered by an unmatched community of frontier labs, big tech, startups, universities, and non‑profits. If we don’t, we’ll be forced”” / X https://x.com/ClementDelangue/status/1948037061304356901
top4 open models are from China, great job brothers!”” / X https://x.com/bigeagle_xd/status/1946426600838586476
US: champion of open markets, ships only closed-source AI. China: master of centralization, ships only open-source AI. Make it make sense.”” / X https://x.com/Yuchenj_UW/status/1947866064500756579
The Open Proof Corpus (OPC) bundles 5,062 human‑checked proofs for 1,010 mathematical competition problems, giving researchers a big public yard‑stick for real reasoning rather than guess‑the‑answer tasks . GEMINI‑2.5‑PRO already judges proofs with 88.1% accuracy, and a simple https://x.com/rohanpaul_ai/status/1948012725122052335
Intelligence isn’t a collection of skills. It’s the efficiency with which you acquire and deploy new skills. It’s an efficiency ratio. And that’s why benchmark scores can be very misleading about the actual intelligence of AI systems.”” / X https://x.com/fchollet/status/1946668452045029861
✅ Try out @Alibaba_Qwen 3 Coder on vLLM nightly with “”qwen3_coder”” tool call parser! Additionally, vLLM offers expert parallelism so you can run this model in flexible configurations where it fits. https://x.com/vllm_project/status/1947780382847603053
🚨 Model Update: Qwen3-coder is in the WebDev Arena! @Alibaba_Qwen have released their best coding model to date and it’s now live in WebDev Arena awaiting your hardest prompts for real world testing. Prompt: “”style a basic login form using Tailwind CSS with dark mode https://x.com/lmarena_ai/status/1948399802947084347
Another incredible OSS model release this summer: the new Qwen 3 update is now live on @togethercompute APi.”” / X https://x.com/vipulved/status/1947871449282216055
Bye Qwen3-235B-A22B, hello Qwen3-235B-A22B-2507! After talking with the community and thinking it through, we decided to stop using hybrid thinking mode. Instead, we’ll train Instruct and Thinking models separately so we can get the best quality possible. Today, we’re releasing https://x.com/Alibaba_Qwen/status/1947344511988076547
Did a benchmark with the new Qwen3 Reasoner 220B on Arena-hard v1 It scores an 89% winrate over gpt4-0314, 4o scores an 81% dont have numbers for o3/4o-mini etc but its basically saturated a near perfect win rate. nicee”” / X https://x.com/Teknium1/status/1948836009183224132
Open source models 📈 qwen3-coder is available in Cline”” / X https://x.com/cline/status/1948452627278430376
Please note, we’re not able to reproduce the 41.8% ARC-AGI-1 score claimed by the latest Qwen 3 release — neither on the public eval set nor on the semi-private set. The numbers we’re seeing are in line with other recent base models. In general, only rely on scores verified by”” / X https://x.com/fchollet/status/1947821353358483547
Qwen just released a 480B coding model & a space to try it out for web dev. Fun! Model: https://x.com/ClementDelangue/status/1947780025886855171
Qwen-MT: Where Speed Meets Smart Translation | Qwen https://qwenlm.github.io/blog/qwen-mt/
RT @Alibaba_Qwen: Performance of Qwen3-Coder-480B-A35B-Instruct on SWE-bench Verified! https://x.com/QuixiAI/status/1947773200953217326
RT @cline: Qwen3-Coder is now available in Cline 🧵 New 480B parameter model with 35B active parameters. > 256K context window > comparabl…”” / X https://x.com/Alibaba_Qwen/status/1947954292738105359
RT @GregKamradt: Anyone have a connection at @Alibaba_Qwen? Trying to reproduce the results on @arcprize and getting different metrics Wa…”” / X https://x.com/clefourrier/status/1947994251410682198
RT @OpenRouterAI: 🟣New: Qwen3-Coder by @Alibaba_Qwen – 480B params (35B active) – Native 256K context length, extrapolates to 1M – Outperf…”” / X https://x.com/huybery/status/1947808085504102487
RT @UnslothAI: @Alibaba_Qwen Congrats guys on another epic release! We’re uploading Dynamic GGUFs, and one with 1M context length so you gu…”” / X https://x.com/QuixiAI/status/1947773516368994320
RT @WolframRvnwlf: I’m now using Qwen3-Coder in Claude Code. Works with any model actually, but this is surely the best one currently. The…”” / X https://x.com/huybery/status/1948184493631959536
We’ve updated Qwen3 and made excellent progress. The non‑reasoning model now delivers significant improvements across a wide range of tasks and many of its capabilities already rival those of reasoning models. It’s truly remarkable, and we hope you enjoy it!”” / X https://x.com/huybery/status/1947345040470380614
Wow the new qwen reasoner at only 232B params is as good as the top closed frontier lab models Big day for OS”” / X https://x.com/Teknium1/status/1948711699013665275
NVIDIA’s Canary-Qwen-2.5B 1st place on the @HuggingFace leaderboard for automatic speech recognition – lowest word error rate (WER) ever recorded on the Hugging Face OpenASR leaderboard: 5.63%. – its the first speech model built on top of an existing LLM. – At its core, it https://x.com/rohanpaul_ai/status/1946823138932863210
AudioRAG is becoming real! Just built a demo with ColQwen-Omni that does semantic search on raw audio, no transcription needed. Drop in a podcast, ask your question, and it finds the exact chunks where it happens. You can also get a written answer. What’s exciting: it skips https://x.com/fdaudens/status/1946226098905169967
so many open LLMs and image LoRAs dropped past week, here’s some picks for you 🫡 LLMs > ByteDance released a bunch of translation models called Seed-X-RM (7B) > NVIDIA released reasoning models of which 32B surpassing the giant Qwen3-235B with cc-by-4.0 license 👏 > LG released https://x.com/mervenoyann/status/1948018642462933149
Looking at the HuggingFace configs, this is a wider/shallower model compared to Qwen3. – 62 layers vs 94 – dim 6144 vs 4096 – 160 experts vs 128 – 96 attn heads vs 64 Curious why the architectural change? Qwen3.5?”” / X https://x.com/nrehiew_/status/1947770826943549732
RT @SIGKITTEN: qwen3-coder, running locally I had it set up testing infra using minunit and gcov and write some tests on a small ~5000 lo…”” / X https://x.com/huybery/status/1948184517673644466
missed this, @NVIDIAAIDev silently dropped Open Reasoning Nemotron models (1.5-32B), SoTA on LiveCodeBench, CC-BY 4.0 licensed 🔥 > 32B competing with Qwen3 235B and DeepSeek R1 > Available across 1.5B, 7B, 14B and 32B size > Supports upto 64K output tokens > Utilises GenSelect https://x.com/reach_vb/status/1947331118983696907
RT @reach_vb: Lets GOOO! @NVIDIAAIDev just dropped Canary Qwen 2.5 – SoTA on Open ASR Leaderboard, CC-BY licensed 🔥 > Works in both ASR an…”” / X https://x.com/reach_vb/status/1946087224346313175
Now it’s possible to do RAG with any-to-any models 🔥 Learn how to search in a video dataset and generate using OmniEmbed, an all modality retriever, and Qwen2.5-Omni, any-to-any model in this notebook 🤝 https://x.com/mervenoyann/status/1947285360926494911
We now have audited data on water consumption for AI. Over the 18 month lifespan of Mistral Large 2, a 128B model, all water usage (including chats, training; hardware & data centers) took as much water as 678 US households use yearly. Each additional query is 45 mL. (Fixed) https://x.com/emollick/status/1947782699948675528
I gave Claude the Mistral report on its AI’s environmental impact and the prompt: “”visualize this in two different ways, one that makes the numbers appear positive, one that makes them seem negative, using vivid comparisons”” (I then had it do some error checking & corrections) https://x.com/emollick/status/1948090558309613587
Yo we made it to #1 yall thanks for checkin out the dataset”” / X https://x.com/Teknium1/status/1946824832764785135
Alibaba launches open-source AI coding model, touted as its most advanced to date | Reuters https://www.reuters.com/world/china/alibaba-launches-open-source-ai-coding-model-touted-its-most-advanced-date-2025-07-23/
Incredible results! Open source is winning. https://x.com/AravSrinivas/status/1947810865685925906
I instinctively believe that the principle of “”denoising””, that is to build examples of structure and order appearing from chaos simply by playing a degradation backward, is extraordinary and can bring AI anywhere.”” / X https://x.com/francoisfleuret/status/1947026623968244008
We are going through what may be the most important period of technological development for mankind, and westerners are blasés about engineering, while Chinese are enthusiastic about it.”” / X https://x.com/francoisfleuret/status/1946817554967626176
Cohere Labs – Catalyst Grants https://cohere.com/research/grants
incredible work on alignment steganography from anthropic fellows i’ve been looking for a straussian explanation of why china keeps publishing open models out of the goodness of their hearts if you do stuff like use open models to, idk, clean *ahem* synthetically paraphrase https://x.com/swyx/status/1947875989666832576
the reason the US is behind on open source models is structural. 1. new labs are not backed by hyperscalers and large vcs. they chose to continue to invest in openai and anthropic instead, which adopted a closed source strategy. deepseek, moonshot, and minimax are amazing. they https://x.com/DaniYogatama/status/1947087827721912485
It’s so beautiful to see the @Kimi_Moonshot team participating in every single community discussions or pull requests on @huggingface (the little blue bubbles on the right). In my opinion, every serious AI organization should dedicate meaningful time and ressources to this https://x.com/ClementDelangue/status/1946208120385999328
When you realize that open-source is at the frontier of AI despite: – less GPUs – less money – less public and policy support – no $100M salaries to attract talent – with closed-source taking advantage and copying all the innovations of open-source without contributing back”” / X https://x.com/ClementDelangue/status/1948756474861084875
Hugging Face bets on cute robots to bring open source AI to life | TechCrunch https://techcrunch.com/podcast/hugging-face-bets-on-cute-robots-to-bring-open-source-ai-to-life/
Robots just got smoother. Real‑Time Chunking (RTC) reduces jerky motion during async inference. Making policies look far more natural. ✅ Applied on Hugging Face’s SmolVLA in the LeRobot repo ✅ Visibly fewer velocity spikes and smoother paths ✅ Works for both diffusion- and https://x.com/IlirAliu_/status/1946613847147618672
Why running local models isn’t just a technical question. @GiadaPistilli in @techreview https://x.com/fdaudens/status/1946300003451859297
New optimized checkpoints of SmolLM3 for faster inference by the PyTorch team 🔥”” / X https://x.com/LoubnaBenAllal1/status/1948477437513208062
Mistral started it DeepSeek scaled it Kimi K2 confirmed it: always more convenient to train an MoE https://x.com/hkproj/status/1947571673021993152
RT @MistralAI: In our continued commitment to open-science, we are releasing the Voxtral Technical Report: https://x.com/andrew_n_carr/status/1947779499032285386
RT @vllm_project: The @huggingface Transformers ↔️ @vllm_project integration just leveled up: Vision-Language Models are now supported out…”” / X https://x.com/ClementDelangue/status/1947775555387916397
The @huggingface Transformers ↔️ @vllm_project integration just leveled up: Vision-Language Models are now supported out of the box! If the model is integrated into Transformers, you can now run it directly with vLLM. https://x.com/vllm_project/status/1947756551663718754
You asked we delivered! Hugging Face Inference Providers is now fully OpenAI client compatible! 🔥 Simply append the provider name to the model ID OpenAI client is arguably the most used client when it comes to LLMs, so getting this right is a big milestone for the team! 🤗 https://x.com/reach_vb/status/1946222374019371437
Here’s why we decided to (1) make Cline open source and (2) not make inference reselling part of our business model: When you control the inference (the AI model calls) and we build the harness (the system directing those calls), neither party can obscure what’s happening. You”” / X https://x.com/cline/status/1946704096888533005
60-second AI video generation just got unlocked! LTXV is the first model to generate native long-form video, with controllability that beats every open source model. – 8× longer than typical gen video – 10–100× faster & cheaper – Runs even on consumer GPUs – Pose, depth & https://x.com/LTX_Video/status/1945465837294440532
This might be the best coding model yet. General-purpose is cool, but if you want the best at coding, specialization wins. No free lunch.”” / X https://x.com/rasbt/status/1947995162782638157
This is the dumbest open source models ever will be”” / X https://x.com/reach_vb/status/1947364340799283539
RT @NetaArt_AI: 🎨 We’re thrilled to officially launch Neta Lumina — the most advanced open-source anime model yet. As our 4th open-source m…”” / X https://x.com/ClementDelangue/status/1947783259028430864
