Special edition — Monday 2 November 2026

Hangzhou-based Deep Principle released Hydrogen-1.0, a scientific model made by post-training Z.ai's GLM-5.3 on chemistry, materials-science and life-sciences data, together with laboratory records and agent trajectories from its own labs. The weights are on Hugging Face under an MIT licence, which lists about 753 billion parameters.

MarkTechPost reports a mixture-of-experts design with a 1-million-token context and 128,000-token maximum output, plus NVFP4 and MXFP4 builds. Citing the company's figures, it reports 38.3% on FrontierScience-Research against 33.3% for GPT-6 Astra, roughly matching Kimi K3 on GPQA-Diamond, while trailing GPT-6 Astra on NMR structure elucidation.

Deep Principle presents it as an open, inspectable alternative to the proprietary models behind recent AI science results. The full training datasets and recipes are not released, according to MarkTechPost.

Anthropic released message threads in public beta on the Claude Platform on Monday. An agent sends the full conversation once to start a thread; after that, each Messages API request carries only the new messages and the ID of the previous response.

Anthropic says threads are already on by default for most Claude Code users, and that on requests of 500,000 to 1 million tokens the median time to first token fell by about 35%. Claude still reads the whole conversation on every turn, so token usage does not change; standard token rates and existing rate limits apply.

Threads are held in memory, encrypted and not written to disk, and expire five minutes after the last request, after which the client resends the full conversation. The beta is offered on the native Claude Platform, Claude Platform on AWS, Claude in Microsoft Foundry and Claude models on Google Cloud.

DeepSeek introduced creator mode for DeepSeek Harness, its open-source agent platform, on Monday, Quartz reported, citing the company's post on X. The Harness page says users can install plugins, or create them through chat in creator mode, to extend the agent's tools, skills and interface.

The company listed a design workspace, a sandbox game and a 3D LLM visualizer as examples. Bloomberg, as cited by Quartz, reported demonstrations of plugins for 3D model generation and a video-editing workbench built from plain-English prompts. DeepSeek says Harness is still in preview, with evolving core plugins and APIs; it can be launched from npm or cloned from its GitHub repository.

It follows the upgraded coding-agent prototype DeepSeek released on Saturday, which this journal covered on 1 November.

Google Threat Intelligence Group published a report on Monday on LLMJacking, in which attackers take over an organisation's access to large language models. Google's Mandiant responded to multiple such incidents in 2026, across North America, South America and Asia Pacific, in financial services, retail, technology and transportation.

In individual incidents, fraudulent charges exceeded $100,000 in a single day, and cumulative charges at some organisations reached $1 million before access was revoked. The main way in was exposed or stolen cloud identity tokens, including service-account keys, IAM credentials and GitHub personal access tokens; some attackers also probed model endpoints left open to the internet.

The report describes a division of labour: some actors gain access while others run LLM mixers, gateways that pool stolen credentials and resell model access. In one case, more than 500,000 API calls in under eight hours powered a routing service called Jet Night, which passed nearly $1 million in token costs to the victim and advertised an unlimited plan for about $60 a month.

SUSE, the open-source enterprise software company owned by private-equity firm EQT, is cutting about 10% of its global workforce, The Register reported, citing an email sent on Monday by chief executive Dirk-Peter van Leeuwen to its roughly 1,900 staff.

In the memo, van Leeuwen wrote that 'automation and AI are fundamentally shifting how software companies operate' and that SUSE's current organisational design and cost base would not carry it forward. Affected staff are to be told by 4 November, and the rest of the company will hear its fiscal 2027 strategy on 16 November. The roles at risk were not specified.

The Register notes that SUSE asked long-serving staff to consider voluntary separation in September.

42AI emerged from stealth with $62.5 million in seed funding and a transformer-based 'behavioral world model' that, it says, estimates the probability of a person's decisions in unfamiliar situations, TechCrunch reported. The company says it works with 10 large brokerages, which use it to manage liquidity, margin requirements and hedging exposure.

Chief technology officer Ehud Ezra said the model draws on signals such as onboarding, support calls, money transfers and trades, including sentiment and voice in customer conversations. Chief executive Amit Gal-Or said revenue is above 'seven figures'.

The seed money came over three years from the founders and Infinity Group, which was founded by Gal-Or's father, Amir Gal-Or, 42AI's chairman; it was not all raised from outside investors.