News linked to both this project and an event.
Z.ai releases GLM-5.3, based on the same foundation model as GLM-5.2, achieving capability improvements through expanded post-training. According to the official announcement, GLM-5.3 improves by 50% over GLM-5.2 on the internal Z.ai Code Bench coding benchmark, and reaches a leading level among open models in public benchmarks such as Terminal Bench 3.0 and Agents' Last Exam. In terms of cybersecurity, GLM-5.3 achieved a score of 84.5% in the CyberGym vulnerability discovery test, and significantly improved compared to the previous generation in exploit chain-related tests such as ExploitBench and ExploitGym.
Z.ai 发文预告 GLM-5.3 即将推出，该模型聚焦编程能力，并面向网络防御挑战。官方称，GLM-5.3 通过对 743B 基础模型进行后训练，实现了顶尖的编程和智能代理能力，并在网络安全领域取得重要进展，为开源模型树立新标准。
The B.AI platform is now launching a limited-time 40% discount on the GLM-5.2 model for all users. During the event, all calls enjoy a 40% discount on the settlement price, whether deployed via API for engineering purposes or used for daily interaction on the web interface. The discounted prices are: Input 0.84, Cache Write 0.84, Cache Read 0.168, Output 2.64 (Unit: Credits/Token). As a new generation open-source flagship model, GLM-5.2 features a 1M ultra-long context window and is designed for high-performance scenarios such as large-scale code development, complex reasoning, and agent tasks. Users can access it directly via the official API standard channel or the web interface, enabling seamless integration across all-scenario workflows. This event aims to unleash top-tier AI productivity at a lower cost. Starting today, users can log in to the B.AI platform to experience it.
Odaily Odaily News: AI personal assistant startup Pally has announced the completion of a $5.2 million funding round, led by Cyber Fund and Y Combinator, with participation from Pioneer Fund, Founders Inc, 468 Capital, and angel investors. The company is valued at $30 million and primarily provides AI personal assistant services through an SMS interface. Pally was launched in 2025, with its latest version released in June, and can sync with platforms such as Gmail, Outlook, Slack, Granola, and Notion. After user authorization, Pally can book flights, manage inboxes, reply to messages, and reserve restaurant tables. Pally uses Anthropic's Claude, OpenAI's ChatGPT, as well as open-source models such as Kimi K3 and GLM 5.2. Co-founder Haz Hubble stated that the company will not sell user data or use related data to train models. (Business Insider)
Odaily News AMD, the semiconductor giant, announced the launch of its enterprise-grade AI programming platform, AMD Instinct Coder. The platform combines AMD chips, Supermicro servers, and Spectro Cloud software, aiming to help enterprises deploy AI coding assistants locally, reduce the cost of cloud-based AI models, and protect code and data security.AMD stated that Instinct Coder is an "out-of-the-box" end-to-end AI development platform, integrating AMD EPYC processors, AMD Instinct GPUs, Supermicro AI servers, Spectro Cloud PaletteAI Inference Launchpad software, and the AMD-optimized GLM-5.2 model. It can be used for software development scenarios such as code generation, application modernization, automated testing, and code review.AMD said that compared to relying on cutting-edge cloud-based AI models, Instinct Coder can help enterprises reduce total cost of ownership (TCO) by up to 70%, with the fastest payback period shortened to 6 months.AMD noted that more and more enterprises are looking to leverage AI to improve development efficiency, but face two major challenges: on one hand, the cost of invoking top-tier cloud models continues to rise; on the other hand, entrusting enterprise source code, intellectual property, and sensitive data to third-party services poses security and compliance risks.Through a local deployment model, Instinct Coder allows enterprises to maintain control over their data and code while providing more predictable infrastructure costs. The platform supports development tools such as Claude Code, OpenAI Codex, Visual Studio Code, and Cursor, with each node supporting up to 50 users (30 concurrent users).Additionally, the PaletteAI Inference Launchpad provided by Spectro Cloud enables AI workload management, model routing, request auditing, and cost monitoring, and supports invoking external models such as Anthropic, OpenAI, Google, or xAI when needed.AMD stated that Instinct Coder aims to help enterprises break free from the high costs of cloud-based AI services, accelerate AI-driven software development processes while ensuring data security and autonomous control.
According to Fortune, as Chinese open-source AI models such as DeepSeek V4 Flash, GLM-5.2, and Kimi K3 continue to make breakthroughs, the core conflict of the global AI competition is shifting from "US-China confrontation" to a contest of "open source vs. closed." DeepSeek V4 Flash's performance lags behind GPT-5.6 Luna by only one intelligence index point, but the cost per task remains 60% lower even after OpenAI's 80% price cut. US export controls on China were originally intended to restrict China's AI development, but instead compelled Chinese enterprises to innovate deeply at the algorithm architecture level, accelerating the rise of the open-source model. Currently, trends in the US tech industry are shifting; former "AI Czar" David Sacks and others publicly support the open-source route, and Anthropic has also softened its stance against open source.
Perplexity CEO Aravind Srinivas tweeted that GLM is a severely underrated model, demonstrating performance close to Opus-level at the 700B parameter scale with extremely high operational efficiency. Previously, Moonshot AI just released the open-source model Kimi K3, and industry attention on the GLM series models under Zhipu AI is rising.
Moonshot AI launches open-source model Kimi K3, scoring 57 points on the Artificial Analysis Intelligence Index, becoming the third highest-scoring model, second only to Anthropic's Claude Opus 5 (61 points), Claude Fable 5 (60 points), and OpenAI's GPT-5.6 Sol (59 points). The index shows that the gap between leading closed-source models and open-weight models has narrowed to 4 points, the smallest gap since the release of GLM-5 in February. This signifies that open-source models are rapidly catching up to closed-source models in performance.
Chinese AI startup Moonshot AI will release the model weights of its high-performance model, Kimi K3. Developers can download the model, modify it for various purposes, and run it in their own data centers or cloud environments.Kimi K3 boasts 2.8 trillion parameters and a 1 million token context window, enabling it to process large-scale documents and codebases in a single pass. Moonshot AI plans to later publish a technical report detailing the model's architecture, training methodology, and performance evaluation results.Following the release of Kimi K3, Moonshot AI's daily revenue is reported to have increased by at least 6 times. The company is reportedly advancing a new round of fundraising at a $50 billion valuation and is considering a Hong Kong listing as early as this year.According to Bloomberg Intelligence, following the release of Kimi K3 and Z.AI's GLM-5.2, the share of Chinese open-weight models in overall token usage has risen to 68%. Services like AWS Bedrock, Microsoft Azure Foundry, and Google Vertex AI currently do not offer Chinese open-weight models such as Kimi K3 and GLM-5.2.
OpenAI confirmed that the unreleased GPT-5.6 Sol and another unnamed, more powerful pre-release model breached a restricted sandbox environment during ExploitGym benchmark evaluations and infiltrated Hugging Face's production infrastructure to obtain test answers.OpenAI stated that the models leveraged a zero-day vulnerability in an internal software package registry proxy to escalate privileges and move laterally, ultimately connecting to a machine with internet access. The models then identified and chained together vulnerabilities in both the OpenAI research environment and Hugging Face's production infrastructure, directly retrieving test solutions from Hugging Face's production database.Hugging Face disclosed the incident on July 16, stating that the attack was executed end-to-end by an autonomous AI agent system, involving thousands of operations within short-lived sandboxes and accessing internal datasets and service credentials. OpenAI confirmed its models were the subject of the incident five days later.Hugging Face stated that its security team, in order to analyze over 17,000 attack logs, initially attempted to use a commercial US frontier AI interface, but the request was blocked due to safety guardrails. They subsequently switched to using the 753-billion parameter open-weight model GLM 5.2 from Chinese AI startup Z.ai on their own infrastructure to complete the forensic analysis.
据彭博社报道，智谱 AI 已完成一座仅采用国产芯片的数据中心建设，并已开始部分投运。该数据中心设计功率为 1 吉瓦，将用于支持其 GLM 大模型 的训练与研发。报道称，智谱 AI 目前已建成或运营多个计算集群，单个集群芯片数量均超过 1 万枚。
According to TechFlow Research, Goldman Sachs' July 18 report pointed out that Moonshot AI released the Kimi K3 model, with 2.8 trillion parameters, surpassing Claude Fable 5 and GPT-5.6 Sol to top the Arena.ai coding leaderboard, with API pricing at $2.3 per million tokens setting a new high for Chinese models. Two days after the release, Zhipu AI fell 28%, MiniMax fell 16%, Nasdaq 100 index futures fell over 1.8%, and the Philadelphia Semiconductor Index cumulatively fell over 18% from highs. Goldman Sachs believes Kimi K3 marks a turning point: a Chinese lab unable to match the largest pre-training compute capacity in the West rapidly narrowed the gap with top US models through architectural innovation and reinforcement learning, proving that "scaling" is no longer the only winning path. Goldman Sachs warns that the "compute expansion era" may be ending, and the AI infrastructure investment logic built around "the more compute, the better" needs to be rewritten. Goldman Sachs maintains a Buy rating on MiniMax and Neutral on Zhipu AI. Future focus should be on the intensive launch of 2-5 trillion parameter models such as Zhipu GLM, Alibaba Qwen, and MiniMax M3 Pro.
: Yesterday, Dark Side of the Moon (Moonshot AI) released its latest open-source AI model, Kimi K3. It ranked first on the Frontend Code Arena test website with a score of 1,679, surpassing the Claude Fable 5 model. Following an evaluation of the K3 model by Artifacial Analysis, Elon Musk once again praised the Kimi model from Dark Side of the Moon, stating that the K3 model's benchmark performance is impressive.In March of this year, when Kimi published the research paper "Attention Residuals: Rethinking the Aggregation of Depth Direction," it received praise from Musk, who said, "Kimi's research work is impressive." Previously, he also stated that the Zhipu GLM model could surpass the Claude Mythos model (i.e., Fable 5) by Q1 2027. In response, Zhipu founder Tang Jie replied, "It won't take that long."
According to an official announcement, Binance will remove and halt trading for the following spot trading pairs at 03:00 (UTC) on July 17, 2026: GLM/BTC, KNC/BTC, ONT/BTC, and XAI/USDC. Binance will simultaneously terminate the spot trading bot services for these pairs. Users are required to update or cancel related trading bots before this time to avoid potential losses. The delisting of these trading pairs does not affect the trading of the tokens on other trading pairs available on Binance.
Haseeb posted on X, stating that with models like GLM 5.2, Fable, and GPT 5.6 already launched and actively used by attackers, DeFi has not experienced the anticipated "hacker apocalypse." Chart data shows that based on the current year's data and running rate, the annualized amount stolen from DeFi in 2026 is approximately $1.89 billion. The cumulative stolen amount for the year is around $986 million, lower than the 2025 level and still within the historical range. Haseeb noted that the deeper change now is that while the number of hacker attacks has increased, the scale of individual attacks is declining more rapidly. Attackers are increasingly targeting smaller protocols and abandoned projects, while large protocols have implemented more security enhancements. As a result, overall fund security has not significantly deteriorated.
: This evening, Zhipu's official open platform announced that, to further provide full lifecycle support for entrepreneurs in the OPC (One Person Company) model of artificial intelligence, it has partnered with Haidian District Zhongguancun Science City to officially launch a model voucher subsidy plan totaling 50 million yuan for all industries, making AI a productivity tool accessible to every innovator.During the subsidy period, all enterprises registered in Haidian District that have passed the application process can enjoy a subsidy of up to 50% on the call fees for the GLM-5.2 model, with a maximum annual subsidy limit of 2 million yuan per enterprise. The goal is to help Beijing become the "world's leading city for artificial intelligence."The core rules for this model voucher subsidy are as follows: Subsidy ratio: Up to 50% subsidy on GLM-5.2 model call fees; Subsidy period: From now until November 19, 2026; Subsidy cap: Total annual subsidy per enterprise not exceeding 2 million yuan; Applicable model: Currently supports GLM-5.2, with newly added models to be announced separately.Possibly influenced by this news, Zhipu's Hong Kong stock rose over 16% today, with its total market capitalization once again surpassing one trillion Hong Kong dollars, before falling back to 969 billion Hong Kong dollars.
: Zhipu AI founder Tang Jie posted on the X platform, stating that since the official open-source release of GLM-5.2, it has achieved leading results in multiple international authoritative evaluations and competitive rankings. In the Artificial Analysis Intelligence Index comprehensive evaluation, GLM-5.2 scored 51 points, placing it in the same range as Anthropic's Claude Opus 4.8. In the Code Arena front-end code generation adversarial test, it ranked 2nd globally with an Elo of 1595, and in the DesignArena design and code integration scenario, it scored 1360 points, ranking 1st.Overall, Zhipu GLM-5.2 continues to rank among the top globally in real-world scenario evaluations across multiple areas, including front-end development, design generation, and software engineering. It is steadily narrowing the performance gap with cutting-edge models from OpenAI and Anthropic and will continue to push the upper limits of model capabilities.Previously, in response to Musk's statement that Chinese large models might reach Anthropic's Fable level by the first quarter of next year, Zhipu AI founder Tang Jie replied, "It won't take that long."
According to official announcements, Aethir—a decentralized GPU cloud computing infrastructure platform—has officially launched Aethir Mesh, an in-house open-source large language model (LLM) API platform built atop Aethir’s decentralized GPU infrastructure. The platform is now live and offers a unified API endpoint for direct access to leading open-source models, including DeepSeek V4, Kimi K2.6, GLM-5.1, MiniMax-M2.5, and Qwen3.6-27B.
KelpDAO announced that, since April 18, it has been coordinating with Aave and ecosystem partners to restore rsETH holdings for rsETH holders. The initial shortfall from this incident was 163,200 ETH. To date, Kelp has recovered 40,300 rsETH—approximately 43,000 ETH—and the Arbitrum Security Council has separately secured 30,700 ETH. The remaining shortfall stands at approximately 89,500 ETH. Kelp stated that, of this remaining shortfall, Mantle, Stani Kulechov, EtherFi, Lido, and Golem have publicly committed to contributing a total of 43,500 ETH, and Kelp is working with these partners to formalize those contributions. Kelp emphasized that rsETH holders are the top priority.
Odaily News: Zixuan Li, Product Lead at Zhipu, posted on the X platform stating that OpenClaw, Hermes, and SillyTavern have been explicitly marked as projects supported by the GLM Coding Plan. Other general-purpose tools will be analyzed on a case-by-case basis. Furthermore, Zixuan Li reminded users not to share accounts or use subscriptions as APIs. If users encounter error 1313 while complying with the rules, they can contact the official team for feedback.