Tech Meridian ← ENTITY INDEX
PROMY MERIDIAN RU

COMPANY · ENTITY #11016

GLM-5.3

Related event timeline, sources and context from the news index.

EVENT TIMELINE

14

RESEARCH · 1 SOURCE · arXiv cs.AI

EvoX: Rust-based multi-model agent runtime with 'Second Policy' and Contrast (arXiv:2611.15006v1)

EvoX is a Rust-based, multi-model agent runtime proposing The Second Policy—treating the harness as an operational policy complementary to a model's semantic policy—and combining Action Realization, Execution Continuity, and Outcome Grounding. Evaluated with the Contrast framework across five benchmarks, six foundation models, and six harnesses, EvoX achieved a 56.2% success rate over 691 task–model pairs, outperforming the average of the other harnesses by ≈3.0 percentage points and showing larger gains on GLM-5.3 (+10.7) and Kimi K3 (+7.9); results were obtained without memory, evolution, or extensions.

6.0

MODELS · 1 SOURCE · NVIDIA Developer

NVIDIA tutorial: serving GLM-5.3 across two DGX Stations with vLLM

NVIDIA published a tutorial demonstrating how to serve the 753B-parameter GLM-5.3 (NVFP4 quantized MoE model) locally across two DGX Stations with GB300 GPUs using vLLM. The guide shows a TP=2 tensor-parallel setup over ConnectX-8 RDMA (no CPU weight offload), with Station A hosting an OpenAI-compatible API endpoint and Station B acting as a worker, and includes steps for checkpoint transfer (≈464 GB), networking checks, and baseline load measurements.

6.0

CODING · 1 SOURCE · NVIDIA Developer

NVIDIA open-sources Snapshot to checkpoint and restore GPU workloads on Kubernetes

NVIDIA released Snapshot, an open-source component of its DSX AI Factory Platform that checkpoints fully initialized GPU workloads (CPU and GPU memory) and restores them on compatible nodes, enabling inference replicas to come online in seconds instead of minutes. Benchmarks and design partners (Crusoe, Saturn Cloud) show restores under 45 seconds and 8x–36x speedups versus cold starts across vLLM, SGLang, and TensorRT LLM setups on HGX B300 hardware.

7.0

MODELS · 1 SOURCE · The Decoder

Tencent AI chief warns of bioweapons risk as Tencent readies open-weight Hy4

Tencent chief AI scientist Yao Shunyu warned that AI will soon be capable of developing bioweapons and argued for a middle ground between fully open and fully closed models. Tencent plans to publish its new model Hy4 with open weights and an accompanying safety report, saying Hy4's biological capabilities fall below a high-risk threshold; the article also highlights attacks on safety filters (e.g., Abliteration techniques, Hugging Face forensic use of GLM-5.2) and gaps in international safeguards.

7.0

MODELS · 1 SOURCE · The Decoder

Nous Research fine-tunes Alibaba Qwen3.8-27B into Arete-27B, cuts pro‑Beijing propaganda to <1%

Nous Research supervised-fine-tuned Alibaba's Qwen3.8-27B on roughly 53 million tokens to produce Arete-27B and says the rate of responses matching Chinese state narratives fell from 22% to under 1% while general capabilities were retained; the team published the model and training data and evaluated it on a custom 803-prompt testbed. The project used Inkling as a teacher model, judged outputs with Grok 4.7, and reports the biggest improvements in free-speech prompts.

8.0

RESEARCH · 1 SOURCE · The Decoder

CMU-led 'Into the Unknown' benchmark finds nine AI agents fall short on open research tasks

Researchers at Carnegie Mellon University and US Department of Energy national labs created the 'Into the Unknown' benchmark to task nine current AI agents with 15 follow-up research problems from national-lab scientists. None of the 135 submissions met all expert requirements: 114 failed integrity checks (fabricated or contradictory claims), only three agents produced submissions without integrity violations (GPT-6 Astra, GPT-5.6 Sol, Claude Opus 5.5), and just one submission—from GPT-6 Astra—was complete; no agent cleared the experts' bar.

8.0

MODELS · 5 SOURCES · Mistral AI · WIRED AI · Hugging Face · The Decoder · TechCrunch AI

Mistral releases Mistral Large 4 (Le Chonk) weights — 1T-parameter open multimodal model

Mistral released the weights for Mistral Large 4 (Le Chonk), a 1‑trillion‑parameter natively multimodal model, claiming leading open‑weight results on coding and defensive cybersecurity benchmarks and new three-mode reasoning options to trade off cost and performance. The company says the model was trained on European infrastructure, is fully open‑weight, shows gains on FrontierCode (19.7%) and artificial analysis indexes (coding 48%, cyber 51%), and achieves a 96.6% average refusal rate on malicious cyber prompts while supporting self-hosted deployment.

8.0

MODELS · 1 SOURCE · The Decoder

Cloudflare previews Clairic, a document-extraction model for Workers AI

Cloudflare released Clairic in preview on Workers AI, an AI model that extracts named fields from documents in a single pass (returning values, formats and confidence scores) and which the company says is faster and cheaper than using LLMs; open weights are slated to follow. Cloudflare notes Clairic can perform simple calculations and forecasts but currently isn't reliable for longer calculation chains and is intended to run in chains alongside other System One models like Jev, Clef and GLM-5.3.

7.0

COMPANIES · 1 SOURCE · TechCrunch AI

Hackers breached OpenCode Console, exposed AI API keys and caused unauthorized charges

OpenCode said attackers broke into an older OpenCode Console system, exposing API keys for about 1.4% of its Console users and triggering unauthorized charges on customers' AI accounts; OpenAI, Anthropic and Google blocked affected keys and OpenAI is assisting refunds. OpenCode's open-source tools were not affected, the company gained access to OpenAI's Daybreak Blue models to aid investigation, and the incident — which coincided with the release of GLM-5.3-Flash — highlighted limits where safety filters can hinder defensive analysis.

7.0

MODELS · 1 SOURCE · The Decoder

Anthropic says new defenses cut Claude reasoning-token theft by up to 99%

Anthropic says it has deployed measures that reduced extraction of Claude's internal reasoning tokens (chain-of-thought steps) by about 90–99% against large-scale distillation attacks. The company says it now restricts access to reasoning steps, runs a classifier to detect extraction attempts and automatically bans flagged accounts (tens of thousands banned), and it named Chinese firms Zhipu and Moonshot as alleged participants whose efforts may have fed models like GLM-5.3; Anthropic warns stolen capabilities could be used in foreign military or surveillance systems and says legitimate users are sometimes affected.

8.0

MODELS · 1 SOURCE · MIT Technology Review AI

Debate over open models intensifies after Anthropic warns GLM-5.3 poses cyberattack risk

MIT Technology Review explains that Anthropic recently released a report arguing GLM-5.3 — a model published in August by Chinese firm Z.ai and available as an open-weight download — poses a major cyberattack risk. The piece frames this claim within the wider public debate between proponents of open models (who emphasize democratization, research access, and lower costs) and safety-focused actors (who warn about misuse), noting trade-offs and differing incentives among companies such as Nvidia and Anthropic.

7.0

COMPANIES · 1 SOURCE · Anthropic

Anthropic says new defenses cut distillation data from Claude by ~90–99%

Anthropic describes a set of expanded defenses to hinder distillation attacks on its Claude models, reporting early evidence that the measures have reduced the amount of data harvested by distillers by roughly 90–99%. The company says it has blocked active extraction tactics, sacrificed revenue and trillions of tokens of paying traffic, and changed model outputs (e.g., summarizing reasoning) to make transcripts less useful for training stolen models, while noting trade-offs for customers and integrations.

8.0

MODELS · 1 SOURCE · The Decoder

AI-powered ARTEX used in likely single-attacker breach of multiple South Korean banks

CrowdStrike reports a suspected Chinese-speaking attacker breached multiple South Korean financial institutions between late September and early October 2026, stealing large volumes of customer data (Shinhan Bank reportedly lost over 25,000 records, per Khan). Researchers say the attacker used ARTEX—an open-source AI automated-penetration-testing tool leveraging DeepSeek v4.1-flash, GLM-5.3, and Grok 4.6—and left Claude Code session logs that indicate attempts to market stolen data.

8.0

MODELS · 1 SOURCE · The Decoder

Anthropic says Zhipu's GLM-5.3 nearly matches Claude Mythos Preview for exploit development

Anthropic's Frontier Red Team reports that Zhipu AI's open-weight model GLM-5.3 can autonomously build cyber exploits close to the capability of Claude Mythos Preview and that GLM-5.3 shipped without effective safeguards. On ExploitBench GLM-5.3 produced working Chrome V8 exploits in 50 of 410 attempts versus Mythos's 56, Anthropic showed the model chained novel vulnerabilities with limited human help, and an independent CAISI assessment reached similar conclusions.

8.0