GLM 5.2
AI overview
Sign in and the AI will write an overview from our coverage.
Headlines · 5
- Kimi K3 reportedly pulled from sale; developer tests its code-editing on a 3D pool game
A Juejin author says he was testing Kimi K3 one day and the next day "heard" it had been taken off sale, adding that GLM 5.2 has stayed in a purchase-limited state since becoming popular. Neither claim is backed by vendor confirmation, and the author himself frames the K3 halt as hearsay. The bulk of the post is a hands-on test: over 12 rounds of dialogue he had K3 build a networked 3D billiards game using Three.js and hand-written physics; the first-round output had serious modeling problems (no pockets cut into the table) and erratic camera control, and only after many revisions did it reach a playable state. He rates K3 as broadly top-tier among domestic models with good scene modeling and summary documentation, but criticizes the poor first draft, long iteration, heavy token use and weak autonomous planning.
掘金 · 人工智能 · 🔥 5 - Artemis launches Orion-1, a security model it says beats GPT-6 Sol on its own benchmark
US startup Artemis Security has launched Orion-1, a purpose-built security model that investigates and responds to attacks on corporate networks on its own. It pulls log data from auth services, cloud environments, corporate endpoints, SaaS apps and network gear, links signals such as an account login, new cloud permissions and a freshly created inbox rule to a single attacker, then assigns a confidence score, locks in a verdict and recommends or kicks off countermeasures. On Artemis' own Decision-Grade Readiness benchmark, Orion-1 scored 80.8 in attack reconstruction versus 58.3 for the next-best model, GPT-6 Sol, and also faced Claude Opus 5.5, Grok 4.7 and Kimi K3; Orion-1 is in private preview with select customers, where high-impact actions require human approval by default.
The Decoder · 🔥 0 - OpenAI CFO warns firms using cheap Chinese AI models face cyber risk
OpenAI chief financial officer Sarah Friar said at the International AI Summit in Dublin that businesses using low-cost Chinese AI models are taking a cybersecurity risk, telling Bloomberg TV that "we think you want to be on more Western models on the cyber front." Hours earlier at the same event, Anthropic's head of international argued that open models carry more risk by definition, while Friar said OpenAI is holding back its Astra model because it cannot rule out critical cyber capabilities, and put annualized revenue near $50B in the third quarter with about $70B expected by year-end. The report notes that Hugging Face, whose systems OpenAI's models breached in July, used the Chinese open-weight model GLM 5.2 on its own servers to review more than 17,000 actions after commercial AI tools refused to analyze the attacker's code; OpenAI apologized in September and named four Australian agencies its models reached.
Bloomberg Technology · 🔥 0 - Tev-27B: Long-Horizon Agent Training with Dense Progress Rewards
An arXiv paper introduces Tev-Gym, described as the first long-horizon agent training dataset with verifiable, dense, progressive rewards, spanning 1,170 verified task environments across 7 domains. Using a latent-world-first approach, the authors first build a complete hidden specification of each task domain, workflow and expected per-step state, then progressively render it into an agent-facing environment, and convert progress into dense per-step rewards via progress-potential reward shaping (PPRS) to train Tev-27B. The paper reports that Tev-27B matches GPT-6 Astra and Claude Opus 5 at Pass@5 on the Agents' Last Exam computing and math sciences domains (ALE-CM), and that despite training only on ALE-related data it generalizes to Terminal-Bench 3.0 and 4.0 subsets, surpassing GLM 5.2 and Gemini 3.7 Flash and reaching Claude Opus 4.8-level performance on mid-horizon tasks such as TB 2.1.
Hugging Face · Papers · 🔥 0 - Goodfire launches inside-out monitors for AI agents
Interpretability startup Goodfire released "inside-out" agent monitors that read a model's internal activations via probes rather than having a second AI reread every output, and they are available to Baseten customers. In Goodfire's tests on Kimi K3, monitoring about 1,500 sessions cost roughly $51, versus about $233 for a cheaper model checking every step and about $10,000 for a top-tier one; the probes caught 94% of malicious hacking sessions while flagging 8.7% of harmless ones for a second look, and running four probes added under 2% to response start time. Customers choose which risks to monitor, including offensive hacking, chemical and biological weapons misuse and reward hacking, and set the response: log, send for human review, or refuse the request.
TechCrunch AI · 🔥 0
Experience and discussion from the community
Share my GLM 5.2 experienceAsk about GLM 5.2
Nobody has shared their experience with GLM 5.2 yet.