Image created with Gemini. Image prompt: Hand-painted Studio Ghibli-style seaside harbor market at golden hour, rows of smiling porcelain masks from varied world festival traditions strung across cobbled stalls, a wooden banner overhead painted with bold clean lettering that reads ‘International’, a traveler in the foreground gently lifting one mask to reveal a hint of soft glowing circuitry beneath, warm terracotta rooftops and teal harbor water, soft gouache watercolor backgrounds, cinematic wide square framing, wistful and quietly mysterious mood.
GLM-5.2 is good but it is not GPT-5.5/Opus 4.8, and even further from Mythos. Yet it is solid & it demonstrates that the open models continue to chase the frontier What is happening is that open weights crossed into GPT-5.2 territory & capabilities at that point are considerable”
https://x.com/emollick/status/2071284966139568440
How good is GLM-5.2? GLM-5.2 (Max) ranks higher than Claude Opus 4.8 (Thinking) on Code Arena, where Frontend coding tasks are being voted on by the community. Below in thread are 10 examples of the same prompts given to both models and completed in a single shot.”
https://x.com/arena/status/2070563149481414779
What are the strategies of Chinese AI companies? To understand this better, @cherylwoooo, @datagenproc, and @ansonwhho scraped >1600 job postings from six major Chinese firms. Here’s what they learned. 🧵”
https://x.com/EpochAIResearch/status/2070190322467144012
The worst-case scenario for the United States is becoming increasingly realistic, and I will briefly explain why. @quxiaoyin raised many valid points, and I agree with her. First of all: -China certainly does not place such strong emphasis on open source because it cares so”
https://x.com/kimmonismus/status/2071524362012791114
UBS says 60% of companies now watching AI budgets are moving to cheaper models and open-source Chinese models The pressure is coming from extreme bills, including users spending up to $35K/month, teams exceeding quotas by 200%, and companies cutting internal AI tools from 5 to”
https://x.com/rohanpaul_ai/status/2070358321232839073
The food delivery leader in China just dropped an open weights 1.6 trillion parameter model while half the USA was sleeping … 🫨 🇨🇳”
https://x.com/JosephJacks_/status/2071858781521342568
[2606.26907] Qwen-Image-Agent: Bridging the Context Gap in Real-World Image Generation
https://arxiv.org/abs/2606.26907
Qwen-Image-Agent Alibaba’s agentic framework bridges the context gap in real-world image generation It plans, reasons, searches, and remembers to build the precise context that text-to-image models need”
https://x.com/HuggingPapers/status/2070489753573548365
WF2026: Software Factories & Keynotes ft. Microsoft, OpenAI, OpenClaw, Z.ai (GLM), MiniMax, HF – YouTube
https://www.youtube.com/live/htM02KMNZnk?t=18296s
omg omg GLM-5.2 is beating fable. china is catching up” chill out and listen to Lisan: > slightly ahead of Opus 4.5 > behind GPT-5.2, Gemini 3 Pro and Opus 4.6″
https://x.com/scaling01/status/2072346101068238946
I can personally attest: OpenClaude using GLM 5.2 is now performing on par with Claude Code powered by Opus 4.8.”
https://x.com/kevincodex/status/2070354383158861955
Introducing ZCode, the official development environment for GLM-5.2 – GLM Coding Plan subscribers: now 1.5x usage quota in ZCode – BYOK supported: works with your existing subscriptions and APIs – Available on macOS, Windows, and Linux Download now:”
https://x.com/Zai_org/status/2072349453361557898
Today’s YouTube video is a longer look at MoA in Hermes Agent, trying to answer some of the common questions: How does using MoA impact cost, speed, and quality? I created an open weights MoA using GLM-5.2, Kimi K2.6 and Minimax M3 to find out, then made a three-headed Grok”
https://x.com/tonbistudio/status/2071595794147250540
𝗚𝗟𝗠-𝟱.𝟮 (the latest open weights model) is having an Enterprise moment, and it is not an exaggeration.🚀 🔥 We have been impressed by how strongly GLM-5.2 is pushing long-horizon performance .. not just in coding, but also in 𝗲𝗻𝘁𝗲𝗿𝗽𝗿𝗶𝘀𝗲 𝗽𝗹𝗮𝗻𝗻𝗶𝗻𝗴, 𝘁𝗼𝗼𝗹”
https://x.com/RajeswarSai/status/2072006835444347390
Got GLM-5.2 running on my Mac Studio via llama.cpp, the reasoning behind all my medical agentic workflows. It orchestrates a swarm of tiny on-device OpenMed experts: oncology, meds, labs. No cloud, no rate limits, nobody can take it away. AI must be owned, not rented.”
https://x.com/MaziyarPanahi/status/2070503452178796704
If I had to choose just one metric, I’d argue that the KV-cache hit rate is the single most important metric for a production-stage AI agent.” – Manus AI prompt caching is important! read about how we do it with deep agents”
https://x.com/hwchase17/status/2070577381392482732
More and more developers turning to glm5.2 as a daily driver for coding Here’s how to use it with deepagents code”
https://x.com/hwchase17/status/2072344890755977571
Built a new agent app for DevRel content research powered by @Zai_org GLM-5.2. Takes your product + audience and turns developer demand into practical blog and talk ideas. Chat-first. Say something like: “I’m working on a Chrome extension for frontend devs that extracts SVGs,”
https://x.com/Astrodevil_/status/2071572680470655253
You can start building with GLM 5.2 in minutes: 1️⃣ Download dcode 2️⃣ Select your model (GLM 5.2) 3️⃣ Enter your API key …and you’re ready to go with frontier performance on open weights. Great demo from @its_ao”
https://x.com/LangChain/status/2072334663457067064
Zai going strong: they officially launched ZCode 3.0. The new AI-native coding IDE is deeply optimized for GLM-5.2 and supports agentic software development from planning and coding to code review and deployment. • Deep GLM-5.2 integration with multi-agent collaboration •”
https://x.com/kimmonismus/status/2072378141041991702
We’ve been impressed with GLM-5.2 and so are introducing a $9.99/month subscription to give you 2-5x discounted access to it and other open weight models like DeepSeek, Kimi, MiniMax, Mimo, Qwen. Use it on Cline CLI & IDE with $1.99 special promo if sign up via: npm i -g cline”
https://x.com/cline/status/2071617325296734309
That Wall Street Journal article about GLM catching up with Mythos (which is not true & the reporting doesn’t back up) is another one of those “everyone will ask me about it at every conference or meeting” articles. Big impact on the policy zeitgeist, even if not fully accurate.”
https://x.com/emollick/status/2071630972521705513
Sonnet 5’s Runescape benchmark scores are big improvement over Sonnet 4.6’s, but outclassed by Pareto frontier neighbors like GLM 5.2 and Gemini 3.5 Flash.”
https://x.com/maxbittker/status/2072054926746779806
Claude Sonnet 5 costs more than Claude Opus 4.8 on the Artificial Analysis Intelligence Index task, and 4.75X more than GLM-5.2. Token efficiency is important.”
https://x.com/Yuchenj_UW/status/2072070274300948497
Sakana Fugu Launches With 93.2 LiveCodeBench Score
https://www.implicator.ai/sakana-fugu-launches-with-93-2-livecodebench-score-after-claude-ban/
Austria urges Europe to host Anthropic following US curbs on AI access | Reuters
https://www.reuters.com/business/austria-lobbies-eu-host-anthropic-ai-after-us-curbs-bloomberg-news-reports-2026-06-28/
Sonnet 5 goes straight into the garbage bin > 1.2x more expensive than Opus 4.8 Max > 2x more expensive than GPT-5.5-xhigh > 5x more expensive than GLM-5.2 > 7x more expensive than Kimi-K2.6 > 57x more expensive than DeepSeek-V4-Pro”
https://x.com/scaling01/status/2072071305281540338
Claude Code Is Quietly Fingerprinting China-Linked API Routers – Vincent Schmalbach
https://www.vincentschmalbach.com/claude-code-china-router-fingerprint/
If you want to read an interesting AI thinking trace, try “I want you to suggest two poems that you think apply very well to the current state of GenAI models like you. Don’t just pick popular poems and back justify. Think hard about options first” in either GLM-5.2 or Opus 4.8″
https://x.com/emollick/status/2070384714691584239
A true survey highlight: World Action Models: A Survey by @NUSingapore World Action Models (WAMs) are embodied systems that connect future prediction directly to action. → WAMs unify Vision-Language-Action policies with predictive world models. This survey covers: WAM”
https://x.com/TheTuringPost/status/2070790639021920356
> the improvement over DFlash: > +20% acceptance length > +14% throughput > avg 127 tok/s vs 111 for DFlash and 81 for EAGLE-3 > for single GPU inference, DSpark’s lightweight + early stopping approach is clearly superior …DeepSeek did just casually ship a SoTA spec decoding”
https://x.com/teortaxesTex/status/2071631028511203457
In their OCR 4 launch this week, Mistral shared a significantly lower score for Chandra 2 than you get from our repo or by running our public code. They also omitted Infinity Parser, which reports 87.6%, from their olmocr comparison.”
https://x.com/VikParuchuri/status/2070465523926630477
The MSA kernel from @MiniMax_AI team is available through the 🤗 Kernels lib and boy, is it speedy 🔥 It currently powers the MiniMax-M3-VL in 🤗 Transformers! Go to [hf dot co / kernels / MiniMaxAI / msa] to get started!”
https://x.com/RisingSayak/status/2072277942554841292
GLM 5.2 just became the first open-source model to lead a category on APEX-SWE. It scored a 55.3% Pass@1 on Integration, the top score we’ve recorded for any model, open or closed source. On the overall leaderboard, GLM 5.2 scored 37.3% Pass@1, ranking 6th place. That makes it”
https://x.com/mercor/status/2072448918751941041
GLM-5.2 is the most intelligent open weights model available, but also the most verbose among the leading models GLM-5.2 (max) used ~141M output tokens (95% reasoning) to run the Artificial Analysis Intelligence Index (1.8x the average model). Key takeaways: ➤ GLM-5.2″
https://x.com/ArtificialAnlys/status/2072022576394821859
South Korea unveils $880bn chip and AI investment plan
https://www.bbc.com/news/articles/c9q2pwzngjqo
This week, during Davos Tech Summit you can see a robotic startup that is tackling Europe’s old infrastructure… Critical infrastructure. From old hydro networks to the world’s shipyards. 60% of it hasn’t been touched in 40 years!!! No way this will not be done by robots! I’ve”
https://x.com/IlirAliu_/status/2072742852166381588
AI concepts and techniques of 2026 ▪️ DeepSeek mHC ▪️ Conditional memory and selective intelligence ▪️ The new fine-tuning stack for LLMs: advanced LoRA + Evolution Strategies ▪️ On-policy self-distillation ▪️ New chips: MatX, Taalas, and the Cracks in the GPU Era ▪️”
https://x.com/TheTuringPost/status/2072483002333200788
We’re about to start seeing feats that are less heroic and more scary. By the fourth week of July 2026, we are likely to see: – DeepSeek-V4-official – Huawei Ascend 950DT SuperPOD (likely beats Meituan’s whole cluster) – ZQ-3 landing (less certain) takeoff begins?”
https://x.com/teortaxesTex/status/2071889274954260720
Huawei Pangu’s 92B 6B active MoE open sourced Big one coming in July They’re basically trying to do Nvidia’s thing with Nemotron”
https://x.com/teortaxesTex/status/2071890951816003663
Underappreciated detail of Meituan claims: as far as I can tell, Yunfan is correct, they run CloudMatrix 384 pods in “910B mode”. So it’s not 50K GPUs, it’s “only” 25K. One 950DT SuperPod with 8192 chips, especially if it’s good for fp8, likely outclasses this rickety system.”
https://x.com/teortaxesTex/status/2071888424823325139
The official GLM-5.2 NVFP4 from NVIDIA is now available. Curious how it compares to other quantizations.”
https://x.com/ZixuanLi_/status/2070391097612783775
GLM-5.2 in NVFP4 is ready to serve in vLLM 🚀 @NVIDIAAI’s official NVFP4 checkpoint of GLM-5.2 on Blackwell cuts the memory footprint vs FP8 while matching its accuracy across reasoning, coding, and long-context benchmarks. Serve it today with: vllm serve nvidia/GLM-5.2-NVFP4″
https://x.com/vllm_project/status/2070569806940848328
First dflash model we trained in house is live (qwen3-32b) and providing ~50% higher throughput for the same hardware. Nice! Need to test dspark next (and then build out dflash or dspark drafters for GLM-5x etc.)”
https://x.com/jon_durbin/status/2072293557172363720
I wonder how much economic value DeepSeek has just created with DSpark. Imagine if *every* Chinese provider improves throughput by 50% 3 months faster. Or improves interactivity by 60%. And anything in between. Billions?”
https://x.com/teortaxesTex/status/2071873225881989424
Why DeepSeek’s DSpark matters: cheaper, faster inference may be the real AI bottleneck 🌟A clear take from Zhihu contributor 平凡 on why DSpark is less about headline model capability and more about making powerful AI usable at scale. TL;DR: DSpark does not make the model”
https://x.com/ZhihuFrontier/status/2071445817102315595
👀 vLLM community is working non-stop to get @deepseek_ai’s new DSpark spec decode algorithm for vLLM! Faster inference for everyone!
https://x.com/vllm_project/status/2071682507775635579
🚀 @deepseek_ai’s DSpark speculative decoding now runs natively in vLLM! What it is: a semi-autoregressive drafter that proposes several tokens in parallel with non-causal sliding-window attention, then verifies them in a single pass. Output stays identical, decoding takes fewer”
https://x.com/vllm_project/status/2072545387639189798
DeepSeek is the forever goat on ML infras. If they haven’t open source all the god-like AI infra projects like DeepSpec / DeepEP / DeepGEMM, we’ll forever be puzzling on why they can achieve super low cost and super high throughput per stream yet still be profitable. As I”
https://x.com/Xianbao_QIAN/status/2071917185380073611
DeepSeek open sources DSpark, a new framework to speed up LLM inference by up to 85% | VentureBeat
https://venturebeat.com/orchestration/deepseek-open-sources-dspark-a-new-framework-to-speed-up-llm-inference-by-up-to-85
GLM 5.2 DSpark preview is here! ✨
https://t.co/DQOMYEiY1o This is the first DSpark speculator for a non-DeepSeek frontier model, trained with Speculators and running on vLLM nightly for ~1.5× faster decode for GLM-5.2-FP8 on 4×B300. Stronger checkpoints to come!”
https://x.com/mgoin_/status/2072525522639212825
new 1.6T total, 48B active open model coming very soon from @Meituan_LongCat custom sparse attention, n-gram, adaptive expert computation, 1M context, 35T of training tokens, muon optimizer, multi teacher OPD, trained on 50k chinese gpus
https://x.com/eliebakouch/status/2071713216028389396
I’ll be in Munich today for EUROPE EMBODIED by @esra_robotics – European Student Robotics Association 🦾 Let’s show everyone what Europe is still capable of! We have a strong research foundation, but is building happening elsewhere? WRONG. Last year, the team around”
https://x.com/IlirAliu_/status/2070416019768656127
The Zhejiang University campus is beautiful! Thanks @thalevinod for the tour. @dolylupec”
https://x.com/TheHumanoidHub/status/2070373306981585366
never thought I’d see natsec cope about a Meituan product. “Bah! Big deal! we have better clusters!” Yes big deal. The whole export control policy, through all its escalations starting with restrictions which resulted in H800 at least, was premised not just on ensuring their”
https://x.com/teortaxesTex/status/2072038240027131963
There’s a lot of sloppy thinking around open models. You can ban them and make it impossible for US companies to use them, but this won’t stop A) global open model progress B) bad actors using them So what exactly is gained by banning open models, including those from China?”
https://x.com/natolambert/status/2070582348203389035
how do these seemingly random Chinese companies have the resources/talent to make their own models?” …”
https://x.com/teortaxesTex/status/2071906284958294419
https://t.co/gPwut02Ilj People are missing out on how big a deal Longcat 2.0 by Meituan (aka “Chinese Doordash”) is. Near frontier performance, trained on 50k Chinese domestic accelerators! The first ever to achieve this!”
https://x.com/sun_hanchi/status/2071664412612833516
Meituan launches LongCat-2.0 1.6T parameter model
https://www.testingcatalog.com/meituan-launches-longcat-2-0-1-6t-parameter-model-on-apis/
now seems like a good day to remind people we have an apache 2.0 coding model you can run with 20 gigs of ram locally for free forever.
https://t.co/verca0bHRC Its a 4 bit quant that gets >99% original performance.”
https://x.com/nickfrosst/status/2070564967279894948
Tour video of our visit to LimX Dynamics (@LimX_Dynamics) in Shenzhen. We saw some impressive demos of Oli and Tron and had a conversation with Paul Li, Head of R&D. We discuss LimX’s tech stack and the fundamental engineering challenges with humanoids. Paul has refreshingly”
https://x.com/TheHumanoidHub/status/2071957152043044963
The talk around openness & AI needs to distinguish between the vibrant and innovative open source movement that is advancing the state-of-the-art on harnesses & other key areas and open weights frontier models, which are entirely dependent on a the goodwill of a few Chinese firms”
https://x.com/emollick/status/2070928706843898117
When we were in China, @xeophon and I made a quick detour to visit Meituan. They continue to be one of our favorite open model builders, as they’re showing how a variety of companies can succeed here and baffle a lot of people as to why they’re making models. Meituan is one of”
https://x.com/natolambert/status/2071972882264268923
Chinese makers have scaled to thousands of affordable humanoid units, but it’s worth noting those are humanoid bodies. Half-decent 5-finger hands with human-like DoF still run $25k to $50k each, and maintenance is quite high. A robot that ships with a well-integrated, reliable”
https://x.com/TheHumanoidHub/status/2072742243287634098
Davos becomes Europe’s first Robot City. Starting today, the Davos Tech Summit turns the entire town into a live testing ground for physical AI: Humanoids, quadrupeds, drones, and other systems working in real-world settings across the city, not on a stage. Speakers include”
https://x.com/IlirAliu_/status/2072605376680280366
Do you think that the auto supply chain and robotics supply chain merger could be a chance for Europe? I mean, we are still strong in quality, precision and complex system integration, right? We do need much faster collaboration between auto suppliers, robotics startups and”
https://x.com/IlirAliu_/status/2071293306789667208
During the AGIBOT tour in Shanghai, we saw an autonomous demo in a controlled setting. The next day, AGIBOT started a 100-hour live stream of the same task on the actual assembly line. Multiple G2 robots are autonomously operating in real tablet production. Today is the day-5″
https://x.com/TheHumanoidHub/status/2070677503770005631
Had a great tour at X-Humanoid in Beijing’s E-Town today. Along with some cool robots and live demos, we got a behind-the-scenes look at their data collection center, where over 150 robots are gathering data.”
https://x.com/TheHumanoidHub/status/2070489791418810709
If you are interested in robotics, and you don’t follow and watch @TheHumanoidHub for his robotics insights (here in China)… no one can help you.”
https://x.com/IlirAliu_/status/2072341681412641141
That’s a wrap on an amazing 17-day journey across China. I didn’t do this alone. My friend @dolylupec came along for the whole ride. She’s got a background in AI and robotics investing, and I’m grateful she chronicled the tours on her account and pitched in on my channel.”
https://x.com/TheHumanoidHub/status/2070873092877091078
The OneRobotics tour was so much fun. The Shenzhen-based company makes cleaning and companion robots for the home, plus a surprisingly fun tennis-playing robot. Their newest is robot Onero, an affordable humanoid, and they build the embodied AI models for all the robots. The”
https://x.com/TheHumanoidHub/status/2072553164080115844
This is the first time I’ve seen a clear blue sky since arriving in China 15 days ago. Beijing is quite nice and it’s bustling!”
https://x.com/TheHumanoidHub/status/2070012061971533825
Tomorrow, Davos becomes Europe’s first Robot City. Starting July 1st, the Davos Tech Summit turns the entire town into a live testing ground for physical AI: Humanoids, quadrupeds, drones, and other systems working in real-world settings across the city, not on a stage.”
https://x.com/IlirAliu_/status/2071865319782805889
Visited DexForce (@DexForceAI) office in Shenzhen on their 5th anniversary and saw some cool autonomous demos. CMO Chris joined us for a short chat about their manipulation-first approach and the deployments they’ve done so far. Thanks to @dolylupec for the camera work.”
https://x.com/TheHumanoidHub/status/2071596919995121766
We visited Astribot (@Astribot_Inc) in Shenzhen, experienced their near-zero-latency teleop, and had a lively chat with Jianan Wang (AI researcher & VP) about their foundation model research and roadmap. Thanks @dolylupec for joining and pitching in!”
https://x.com/TheHumanoidHub/status/2071658106615197805
We visited Paxini in Shenzhen, a company focused on the sense of touch for robots. We got hands-on with their dual-modal Dexter Hands (cameras + tactile sensors) and their Tora Double One humanoid. Humanoid robots will need this touch layer to get truly dexterous. @dolylupec”
https://x.com/TheHumanoidHub/status/2072338013770924468
Half a year ago, a Zhihu contributor predicted that the next Transformer would absorb loops, recurrent state, sparse routing, and latent reasoning. Now, as Loop Engineering gains momentum, that prediction feels strikingly relevant. Let’s dive in 👇 What Will the Next Generation”
https://x.com/ZhihuFrontier/status/2070442689427058900
Leave a Reply