Image created with Ideogram v3. Image prompt: Late‑90s boy‑band cover “World Tour – Global Harmony”: members stand on stylized spinning globe, waving flags; denim and leather mix; night sky with aurora; chrome title with orbiting ring.
“We have optimized the Qwen3 models for coding and agentic capabilities, and also we have strengthened the support of MCP as well. Below we provide examples to show how Qwen3 thinks and interacts with the environment. https://x.com/Alibaba_Qwen/status/1916962100817367192
“Today Anthropic submitted key recommendations on the “Diffusion Rule” – export controls on advanced AI chips. We believe maintaining America’s compute advantage is essential for national security as powerful AI systems develop. https://x.com/jackclarkSF/status/1917629783090831582
“BOOOOM! Qwen 3 235B MoE (22B Active) – beats R1, Grok, O1 AND Apache 2.0 licensed! 🔥 https://x.com/reach_vb/status/1916965315910553886
“Qwen3 exhibits scalable and smooth performance improvements that are directly correlated with the computational reasoning budget allocated. This design enables users to configure task-specific budgets with greater ease, achieving a more optimal balance between cost efficiency and https://x.com/Alibaba_Qwen/status/1916962091925442698
“Introducing Qwen3! We release and open-weight Qwen3, our latest large language models, including 2 MoE models and 6 dense models, ranging from 0.6B to 235B. Our flagship model, Qwen3-235B-A22B, achieves competitive results in benchmark evaluations of coding, math, general https://x.com/Alibaba_Qwen/status/1916962087676612998
“Is Qwen3-235B the new budget-friendly coding champ in Cline? Early user feedback is rolling in — it’s promising, but not perfect. Here’s what we’re hearing from the Cline community: 🧵” / X https://x.com/cline/status/1917708041857949983
Australian radio station secretly used an AI host for six months | The Independent https://www.the-independent.com/tech/ai-radio-host-australia-cada-elevenlabs-b2740033.html
Blog | Localforge | Localforge https://localforge.dev/blog/running-qwen3-macbook-mlx
“Qwen3-235B-A22B Superior to OpenAIs o3-mini in all the benchmarks 👀 Now it’s all about API pricing and further testing” / X https://x.com/scaling01/status/1916967634786029722
“Qwen3 is now out, but how do you run it locally? Spent the day (not knowing Qwen3 was today until an hour ago) getting @huggingface ChatUI + Qwen2.5 72B running 100% locally. Faced some headaches and confusing docs, so distilled it down for you all: https://x.com/TheZachMueller/status/1916969775525191684
“BOOOOM! you can now use the latest DeepSeek Prover V2 directly on the model page powered by @novita_labs 🔥 Open Source FTW! 💥 https://x.com/reach_vb/status/1917549921470972172
“Although I left DeepSeek quite a while ago, being able to scale up to 671B truly feels like a dream come true for me. I’m deeply grateful to ZZ, Zhihong and other colleagues at DeepSeek for their support, to Liang for the opportunity, and to everyone in the field who has” / X https://x.com/huajian_xin/status/1917603640124363090
“We just released DeepSeek-Prover V2. – Solves nearly 90% of miniF2F problems – Significantly improves the SoTA performance on the PutnamBench – Achieves a non-trivial pass rate on AIME 24 & 25 problems in their formal version Github: https://x.com/zhs05232838/status/1917600755936018715
“The long-awaited Qwen3 is finally here! Our team has put tremendous effort into Qwen3, hoping to bring something fresh to the open LLM community. We’ve made significant progress in pretraining, large-scale reinforcement learning, and integration of reasoning modes. We believe” / X https://x.com/huybery/status/1916962562056524177
“We also evaluated the preliminary performance of Qwen3-235B-A22B on the open-source coding agent Openhands. It achieved 34.4% on Swebench-verified, achieving competitive results with fewer parameters! Thanks to @allhands_ai for providing an easy-to-use agent. Both open models and https://x.com/Alibaba_Qwen/status/1917064282552078480
“Alibaba’s Qwen just unveiled Qwen3: a family of eight open models ranging from 600M to 235B params. — Flagship version rivals OpenAI o1 & DeepSeek-R1 — Hybrid “thinking” mode in all models — Boosted coding + agent performance — Supports 119 languages https://x.com/rowancheung/status/1917095301485052142
“Qwen3 and Qwen3 MoEs are already supported in the latest mlx-lm thanks to @Prince_Canuma and @ActuallyIsaak pip install -U mlx-lm Awesome that @Alibaba_Qwen ships a model for every device: -iPhone: 0.6B, 4B -Macbook: 8B, 30B, 3B/30B MoE -M2, M3 Ultra: 22B/235B MoE” / X https://x.com/AwniHannun/status/1916862553852203349
“Qwen3 and Qwen3 MoEs are already supported in the latest mlx-lm thanks to @Prince_Canuma and @ActuallyIsaak pip install -U mlx-lm Awesome that @Alibaba_Qwen ships a model for every device: -iPhone: 0.6B, 4B -Macbook: 8B, 30B, 3B/30B MoE -M2, M3 Ultra: 22B/235B MoE” / X https://x.com/awnihannun/status/1916862553852203349
“The Qwen Chat APP is now available for both iOS and Android users! It’s free to use and designed to assist with creativity, collaboration, and endless possibilities. Just ask, and let Qwen Chat handle the rest. Scan the QR code to quickly access the Qwen Chat APP! https://x.com/Alibaba_Qwen/status/1915761990703697925
“Feel free to download the Qwen Chat Android APP by scanning this QR code! https://x.com/Alibaba_Qwen/status/1915942739855937560
“So, how Qwen3 looks like imo: Main line: finegrained MoE, DeepSeek-like (V2 and V3-Lite scaled), GQA, trained with global-batch load balance, 25T tokens, 256K context, some improved GRPO (DAPO?), unified chat/reasoner, flagship is Sonnet 3.7 tier Dense models largely as before https://x.com/teortaxesTex/status/1916779853111509498
Duolingo Launches 148 New Language Courses – Duolingo, Inc. https://investors.duolingo.com/news-releases/news-release-details/duolingo-launches-148-new-language-courses
Exclusive: Trump officials eye changes to Biden’s AI chip export rule, sources say | Reuters https://www.reuters.com/world/china/trump-officials-eye-changes-bidens-ai-chip-export-rule-sources-say-2025-04-29/
Nvidia CEO Jensen Huang says China ‘not behind’ in AI https://www.cnbc.com/2025/04/30/nvidia-ceo-jensen-huang-says-china-not-behind-in-ai.html
“Qwen3 is a win for open weights & efficiency – hybrid reasoning models that approach DeepSeek R1’s GPQA score with 1/3 the total parameters and a range of smaller models suited for compute limited environments Today, Alibaba announced eight hybrid reasoning models of varying https://x.com/ArtificialAnlys/status/1917246369510879280
DeepSeek available to download again in South Korea after suspension | Reuters https://www.reuters.com/sustainability/boards-policy-regulation/deepseek-available-download-again-south-korea-after-suspension-2025-04-28/
DeepSeek-R2: China’s Powerful New AI Model for 2025 https://deepseek.ai/blog/deepseek-r2-ai-model-launch-2025
China’s Xi calls for self sufficiency in AI development amid U.S. rivalry | Reuters https://www.reuters.com/world/china/chinas-xi-calls-self-sufficiency-ai-development-amid-us-rivalry-2025-04-26/
DeepMind UK staff plan to unionise and challenge deals with Israel links, FT reports | Reuters https://www.reuters.com/sustainability/sustainable-finance-reporting/deepmind-uk-staff-plan-unionise-challenge-deals-with-israel-links-ft-reports-2025-04-26/
Benchmarking LLMs for global health https://research.google/blog/benchmarking-llms-for-global-health/
“The Qwen team really cooked with this release Just incredible work all around: – 235B MoE that is comparable to o1, o3-mini, Gemini 2.5 Pro, etc. – trained on 36T tokens, covering 119 languages! Data extracted from PDFs, synthetic data, etc. – Thinking and non-thinking modes – https://x.com/iScienceLuvr/status/1916966249588002867
“QWEN-3 is finally out! > Matches Gemini 2.5 Pro performance > Outperforms OpenAI o1 > Open-sourced (Apache 2.0) > 119 languages, 32K–128K context https://x.com/LiorOnAI/status/1916998817725223240
“ByteDance’s Dreamina AI officially launched Seedream 3.0, a new text-to-image model —Produces 2K cinematic visuals with realistic textures —Delivers precise typography — #2 on Artificial Analysis’ Image Arena behind only GPT-4o https://x.com/rowancheung/status/1915288450187645016
“Pretty fucking incredible week so far: > Qwen3 – MoE (235B, 30B) + Dense (32, 14, 8, 4, 0.6B) > Xiaomi – MiMo 7B dense > Kyutai – Helium 2B dense > DeepSeek – Prover V2 671B MoE > Qwen2.5 Omni 3B > Microsoft – Phi4 14B Reasoning, Mini (3.8B) & Plus > JetBrains- Mellum 4B Dense” / X https://x.com/reach_vb/status/1917938596465750476
“> Qwen3 drops > 235B total / 22B active > neck to neck with LLaMA-4-Maverick zucc in absolute shambles https://x.com/ns123abc/status/1916971024509280503
“Dynamic Qwen3 GGUFs are here! – Run them in llama.cpp, lmstudio and ollama nowww! 💥 https://x.com/reach_vb/status/1916982114462900726
“Qwen’s distillation teacher MoE has fewer total parameters (235B) than Meta’s Llama 4 Behemoth has active (288B). As a result its still much smaller distills dunk on Scout viciously. Should have trained that behemoth ass to a fit condition first, huh https://x.com/teortaxesTex/status/1916971319800823932
“We shipped an alpha version of the new Surya OCR model. No hype, just facts: – 90+ languages (focus on en, romance langs, zh, ar, ja, ko) – LaTeX and formatting – Char/word/line bboxes – ~500M non-embed params – 10-20 pages/s https://x.com/VikParuchuri/status/1915492483955384659
“Introducing ERNIE X1 Turbo & ERNIE 4.5 Turbo! Building on the success of ERNIE X1 and 4.5, the upgraded ERNIE X1 Turbo and 4.5 Turbo deliver results faster and cheaper. Both models stand out for their multimodal capabilities, strong reasoning and low costs. For X1 Turbo, input https://x.com/Baidu_Inc/status/1915603080336597310
“OpenAI 🤝 Singapore Airlines:” / X https://x.com/gdb/status/1915866162438803629
Qwen 3: 0.6B to 235B MoE full+base models that beat R1 and o1 | AINews https://news.smol.ai/issues/25-04-28-qwen-3
“The chatter is that DeepSeek R2 is going to be released soon…” / X https://x.com/iScienceLuvr/status/1916365312145924532
QwQ-32B claims to match DeepSeek R1-671B | AINews https://news.smol.ai/issues/25-04-16-ainews-qwq-32b-claims-to-match-deepseek-r1-671b
“DeepSeek R1T Chimera – Merge DeepSeek V3 & R1 – 40% fewer tokens, WITHOUT performance loss – MIT licensed 🔥 https://x.com/reach_vb/status/1916490086188736602
“Qwen 3 235B now on @togethercompute API! Qwen 3 is a reasoning model that has a non-reasoning instruct mode with allowance for setting a thinking budget. It’s efficient ($0.20/M input & $0.60/M output on our throughput optimized endpoint) and fantastic on a variety of” / X https://x.com/vipulved/status/1917777842466889873
Qwen 3: The new open standard – by Nathan Lambert https://www.interconnects.ai/p/qwen-3-the-new-open-standard
“Qwen-3-MoE vs DeepSeek V2 (original) their designs are superficially similar – but different This will be a very interesting test of a few scaling laws https://x.com/teortaxesTex/status/1916824004901359943
Qwen3: Think Deeper, Act Faster | Qwen https://qwenlm.github.io/blog/qwen3/
“🎉 Congrats @Alibaba_Qwen on releasing the new Qwen 3 family! – Qwen-3 from 0.6B to 235B all open under Apache 2.0 – Qwen3-235B-A22B competitive against the best proprietary models across hard benchmarks Challenge them with your toughest prompts in the Arena! https://x.com/lmarena_ai/status/1917245472521289815
“first impression is that Qwen 30B-3A will be the star of the show” / X https://x.com/teortaxesTex/status/1916918829050998981
“So Qwen 3-235B with thinking seems good, but not blowing away any of my weird frontier tests, some of which DeepSeek r1 did better. It did okay generating a p5js starship (though it had errors to correct), but failed the Lem Test and couldn’t do a twigl shader in many attempts. https://x.com/emollick/status/1917022882888142926
“⬆️ pip install -U vLLM vllm serve Qwen/Qwen3-235B-A22B-FP8 –enable-reasoning –reasoning-parser deepseek_r1 –tensor-parallel-size 4 vLLM introduce Day 0 support for @Alibaba_Qwen Qwen3 and Qwen3 MoE model architecture. Try it out: https://x.com/vllm_project/status/1917008899410215275
“You can now run inference directly on the Qwen 3 235B Hugging Face model page – powered by Together AI! https://x.com/togethercompute/status/1917616701249565120
“Qwen3 models have a cool feature: toggle thinking mode on and off. It’s a chat template option, so presumably works by including / excluding the `<think>` tokens. (IBMs Granite 3.3 models had a similar feature). Here’s how to use it with mlx-lm: https://x.com/awnihannun/status/1916932256578605246
“Nice UI for managing the thinking/non-thinking modes of Qwen3 https://x.com/fdaudens/status/1916981928009285858
“Qwen3 is out https://x.com/fdaudens/status/1916970577425846446
“🎉 Qwen3 235B is now on HuggingChat! https://x.com/fdaudens/status/1917317723547218352
“Qwen3 is out! SkyPilot is excited to be a close friend with the @Alibaba_Qwen team. Let’s spin up Qwen3 easily on your clusters or clouds with one SkyPilot command! https://x.com/skypilot_org/status/1916987145195295095
“Qwen3-30B-A3B is de facto on par with Qwen3-32B dense and the greatest vindication of finegrained MoEs the world has seen in the open. https://x.com/teortaxesTex/status/1916966009170251899
“Qwen3 models are supporting 119 languages and dialects. This extensive multilingual capability opens up new possibilities for international applications, enabling users worldwide to benefit from the power of these models. https://x.com/Alibaba_Qwen/status/1916962096346202468
“ByteDance just announced QuaDMix on Hugging Face Quality-Diversity Balanced Data Selection for Efficient LLM Pretraining https://x.com/_akhaliq/status/1915656590130036887
