AI Coding Live News

Hangzhou-based AI-for-science company Deep Principle released Hydrogen-1.0 on November 2, its first general-purpose scientific model. It post-trains Zhipu's GLM-5.3 with SFT and RL on chemistry, materials and life-science data plus its own lab records and discovery trajectories, and ships about 753B parameters of BF16 weights on Hugging Face under the MIT license. By the vendor's own evaluation it scores 38.3% on FrontierScience-Research versus 33.3% for GPT-6 Astra; none of the results have been independently reproduced yet.

Zoom

Epoch AI reported on Oct 29 that, in a deliberately basic setup emulating an unsophisticated attacker, Z.ai's open-weight GLM 5.3 read security fixes that were public but not yet shipped and autonomously built exploit chains for all six Firefox releases from Sep 17 to Oct 26, letting a single webpage visit read local files. Marginal cost was about $1,000 per version and the fastest run took 12 hours. Mozilla patched a previously undisclosed bug the model also used in Firefox 159.0 on Oct 27.

Zoom

Shanghai AI Laboratory has moved Atria Dawn from Preview to a full release, with open weights (plus an FP8 build) on Hugging Face under internlm/Atria-Dawn and on ModelScope, MIT licensed. Built on the 744B-parameter MoE GLM-5.2, it targets long-horizon research and engineering agent work; context doubles from the Preview's 256K to 512K and input is text only. The vendor table shows it leading the listed models on AutomationBench, CyberGym and SkillsBench, while still trailing Claude Opus 5 on coding tests such as SWE-bench Pro and Terminal-Bench 2.1 (all vendor-reported).

Zoom

In an Oct 26 incident report on its blog, Cloudflare said that between Aug 27 and Sep 9, 155 responses from GLM-5.3 on Workers AI (@cf/zai-org/glm-5.3) to prompts longer than 681,280 tokens contained data from other customers' recent requests, all returned with HTTP 200. The cause was a prefill KV cache pool smaller than the 1,048,576-token context window, combined with an SGLang bug in disaggregated prefill/decode mode. Cloudflare found no evidence of deliberate exploitation, has fixed it, and recommends operators upgrade to SGLang v0.5.22 or later.

Zoom

On October 21 Letta released FrontierMemory 1.0. Instead of quizzing agents on facts such as birthdays or hobbies, it gives an agent a set of past trajectories, has it write memory or skills, and scores that memory by how many coding and knowledge-work tasks a task agent then solves. Version 1.0 has 15 tasks adapted from public benchmarks such as DeepSWE, SkillsBench and APEX-Agents. On the model board (memory written in Letta Dream, tasks run in Letta Code) Opus 5.5 and GLM-5.3 both score 73.3% and Grok 4.7 scores 33.3%; on the harness board Letta's own Letta Dream ranks first. All numbers are Letta's own.

Zoom

On Oct 20, AI startup humans& released its first model, Ampersand v0.1, pitched as a collaborator that helps groups reach individual and shared goals rather than an assistant that completes tasks alone. It is trained from Z.ai's GLM 5.3 (about 744B total, about 40B active parameters) with long-horizon multi-party RL against simulated users played by humans&'s own user model, Persimmon. It is available now, with no waitlist, through an API compatible with OpenAI's Agents API.

Zoom

On October 19 Cloudflare released a preview of Clairic, a "System One extraction model". Instead of generating tokens, it uses one parallel forward pass to extract, compute, estimate and forecast typed values from text or up to 4 images, with confidence, character spans and a null probability for each answer, returning null when a value is missing. It is live on Workers AI as @cf/cloudflare/clairic; the docs list a 64K context, vision support and $0.10 per million input tokens with no output charge. In Cloudflare's example, a Clairic+Clef+GLM-5.3 cascade ran about 3.2x faster and about 49% cheaper than GLM-5.3 alone (company figures). Open weights are planned later.

Zoom

The SGLang team released v0.5.22 featuring 950 PRs from 245 contributors: DeepSeek-V4.1 delivers up to 2.3x more throughput per GPU on NVIDIA H200 and 1.9x on AMD MI350X/MI355X via DSpark. The new InstantTensor loader cuts model loading time by 10x. B200 decodes Qwen3.8-Flash-Next 12% faster, while B300 speeds up GLM-5.3-Flash by 10% and yields 1.85x decode throughput for MiMo-V2.6-Pro. SGLang also adds native Apple Silicon PyTorch support and ComfyUI integration.

Zoom

Can LLMs reliably maintain spatial state representations across sequential operations? USC, UCLA, and CMU introduced Grid-Bench (arXiv:2610.22312, GitHub: grid-bench/Grid-Bench), an executable benchmark with 8,928 instances testing 2D/3D/4D grid state transitions across history lengths up to T=64. Evaluations across 6 frontier models show steep degradation: while GPT-5.6 Sol and Claude 5 Sonnet retain strong cell accuracy, open-weight and mid-tier models collapse to near-zero exact world reconstruction over long trajectories.

Zoom

vLLM shipped v0.32.0 on Oct 18 with 814 commits from 363 contributors (121 new). Highlights: a default-served POST /v1/systemone structured-decision endpoint and a generic hf tool/reasoning parser; NVFP4 KV cache on SM8x/SM90/SM12x and FP8 KV by default for GLM-5.x on SM100; Model Runner V2 gains a GPU n-gram speculator and Elastic EP and becomes the ROCm default; three security advisories are fixed, default CORS is restricted, and Python 3.10 is dropped.

Zoom

vLLM and NVIDIA unveiled a unified CuTe DSL decode megakernel (~12,000 LoC authored by NVIDIA CentML's Benjamin Chislett) for Moonshot AI's frontier Kimi K3 model on NVIDIA GB300 NVL72. By fusing attention, MoE, and cross-GPU communication, the optimization cuts per-layer latency from 130 µs (23 kernels) to 69 µs (13 kernels), delivering 842 tokens/s at batch=1 with DSpark speculative decoding (1.73x faster than stock vLLM) and up to 1.8x higher P90 per-user interactivity (403.2–423 tok/s/user) on SemiAnalysis AgentX, with DeepSeek-V4.1-Flash (PR #61868) and GLM-5.3 support coming next.

Zoom

Ollama launched US- and EU-hosted Ollama Cloud with strict zero data retention and zero training, adding immediate support for DeepSeek-V4.1-Flash-1017 via one-line launch commands for Claude Code and OpenCode alongside Ollama v0.40.6. Meanwhile, SemiAnalysis revealed that GLM-5.3 was built entirely by post-training the previous GLM-5.2 base model without pretraining from scratch, achieving a 'night and day difference' in cybersecurity and agentic capabilities.

Zoom

Addressing KV cache bottlenecks in long-context LLM serving, a new production study on 8xB300 nodes serving GLM-5.3 (arXiv:2610.21500) reveals that unsharded remote KV caches inflate hit-heavy p99 TTFT to 83.32s, and resolves it via rank-sharded asynchronous LMCache loading—boosting GET bandwidth 6.3x to 10.76 GB/s and slashing p99 to 0.67s (+34% throughput). Meanwhile, Baidu PaddlePaddle open-sources Virtual Head Attention (VHA, arXiv:2610.17959), folding linear Q-Premix and Postmix into standard GQA-2 weights at inference to halve KV cache while beating MLA (+0.70%) and GQA-4 (+2.60%) on a 30B-A3B MoE and lifting 8K-128K RULER from 73.0 to 84.4.

Zoom

Advancing autonomous AI-for-AI (AI4AI) engineering, researchers from Shanghai AI Lab (InternScience) release MLExpert (arXiv:2610.22344), a 35B Mixture-of-Experts agentic model for autonomous machine learning engineering. Powered by the Kaggle-MLE Factory, hierarchical agentic SFT, and Windowed Interaction RL (WIRL), MLExpert boosts average medal rates on MLE-Bench Lite from 37.9% to 72.7%—matching frontier giants like Kimi-K3 and GLM-5.3 while generalizing to agent scaffold optimization.

Zoom