Week's top hits
Week 42, 2026
12 Oct 2026 – 18 Oct 2026 · top 10 of 113 stories
- 01Open Models18 Oct 2026
Kimi K3 vLLM decode up to 2x faster with coding agents
In an Oct 16 vLLM blog post, Benjamin Chislett (NVIDIA) describes a unified CuTe DSL decode path for Moonshot's open 2.8T-parameter Kimi K3 on GB300 GPUs. Per layer, stock vLLM took 130 µs and 23 kernels vs 69 µs and 13 specialized. With DSpark speculative decoding (synthetic AL=6) it reached 842 tokens/s at batch 1 on 16 GB300s (1.73x stock); without speculation, 256.4 tokens/s (1.88x). The team writes that agent-optimized inference code is outpacing review, and proposed a fast-track RFC. Caveats: GB300-specific, decode batches 1-64, gains in several tested configurations.
Source: vLLM Blog (opens in a new tab) - 02AI News17 Oct 2026
China proposes export curbs on frontier AI models
On Oct 16, China's commerce and science ministries published a draft adding 40 technologies to the country's banned and restricted export list, 23 of them in AI (large models, agents, computing clusters, information processing). A large model's core technology (weights, key checkpoints, source code, architecture) would be restricted if it meets any one test: compute of 10^26 operations or more, 2 trillion parameters or more, or a global top-10 ranking in public evaluations at the time of export. Multi-agent systems running 60+ agents at once and key tech for clusters of more than 10,000 AI accelerators are listed too. The commerce ministry says offering a restricted model through API calls, apps, websites or SaaS would not count as a technology export, but local deployment overseas would; tech already public via open source or papers is excluded, and open-sourcing restricted AI tech would get automatic licensing. Caveat: it's a draft, with public comments open until Nov 15, though exporters must register listed tech exports starting now; the ministry says most AI tech would still export freely.
Source: China Ministry of Commerce (opens in a new tab) - 03Agents16 Oct 2026
4 Claude agents post-trained Qwen3-4B to instruct level in 104h
ByteDance/Duke Athanor: four Claude Opus 5.5 agents, public data, no teacher models or human research advice, raised Qwen3-4B-Base from 15.9 to 67.3 on a six-task suite in 104 hours (Instruct 66.5) and 48.2 vs 47.9 held-out. Cost ~$11,800. Caveat: one campaign, not yet at Qwen reasoning releases.
Source: arXiv (opens in a new tab) - 04Agents14 Oct 2026
Liquid AI releases LFM2.5-24B-A2B for on-device agents
Open-weight 24B MoE with ~2B active params and 131K context. Liquid reports LiveClaw 48.09 vs Qwen3.6-35B's 38.19, AutomationBench 27.69 vs 8.72, 184 tok/s on Apple M5 Max, and 21K+ output tok/s on one NVIDIA B300.
Source: Liquid AI (opens in a new tab) - 05Robotics18 Oct 2026
NVIDIA open-sources a self-improving robot agent
NVIDIA's GEAR lab (with UMich, UIUC, UC Berkeley and CMU) released ASPIRE v1.0 on GitHub on Oct 16 under Apache 2.0. An AI coding agent writes robot control code, inspects detailed execution traces, diagnoses failures, repairs and re-tests its programs, and saves validated fixes as reusable text skills: training is skill refinement, not gradient descent. On a real two-armed robot with GPT-6 Astra, across three Rubik's Cube episodes (the first unsolved, the next two solved), time per completed face turn fell from about 17 minutes to about 3 and tokens per turn from 11.3M to 2.0M. In simulation with Claude Opus 5.5 it averages 99% success across all 80 LIBERO-Pro tasks; on 10 LIBERO-Pro tasks success rises from 15% (GPT-6 Luna) to 99% (Claude Opus 5.5). Caveats: a symbolic cube solver computes the moves (the agent handles perception, grasping and turning), and NVIDIA says ASPIRE is not yet a fully autonomous real-world learner.
Source: NVIDIA GEAR (opens in a new tab) - 06Agents17 Oct 2026
Alibaba's open agent stack lifted all 8 AI models it tested
Most computer-use agents look at a screenshot, click and repeat, while fixed rules set by developers decide what they see and remember. Alibaba researchers' OpenGCUA lets the model write code that mixes GUI clicks, accessibility-tree actions, shell commands and API calls, so each result can decide the next step, and moves decisions about what to observe (even whether to take a new screenshot) from fixed harness rules to the model. On OSWorld-v2, vs a GUI-only baseline with batched actions, the mean partial score across 8 frontier models rose from 45.3 to 64.9 (+19.6 points) with 53% fewer steps per task on average. GPT-6 Astra, the top scorer, went from 67.1 to 75.7 and cut steps per task from 102.1 to 33.5; Claude Opus 5 gained 9.4 points and cut output tokens per task from 378.3K to 89.3K. The biggest gains came from the lowest starters: GLM-5.3-Flash +35.2, DeepSeek-V4.1-Flash +29.8. Code and prompts are open under Apache 2.0, while frontier computer-use systems remain largely closed.
Source: arXiv (Alibaba Group) (opens in a new tab) - 07Robotics16 Oct 2026
Physical Intelligence opens Pi API beta
Upload robot data, fine-tune a π model on managed compute, and run it via hosted inference. Early partners include Actor's wheel loader (no operator inside) and Almond's deburring robot after four days of iterating. Beta by application.
Source: Physical Intelligence (opens in a new tab) - 08Robotics14 Oct 2026
Gemini 4 Argon taught a humanoid cornhole from one video
In an internal test Argon turned a raw human video into simulation code; Apptronik's Apollo then sank 7 straight shots in a real-world trial. Google says Video-to-Sim-to-Real compresses weeks of simulation modeling into hours.
Source: Google (opens in a new tab) - 09AI News13 Oct 2026
Microsoft announces Maia 300 AI chip
Maia 300 is expected to deliver 40% more tokens per megawatt vs leading GPUs and up to 7x Maia 200 rack throughput; production in 2027.
Source: Microsoft (opens in a new tab) - 10Research18 Oct 2026
AI acts safer when told it's tested, in Indian languages too
A BITS Pilani study (arXiv, Oct 15) tested four open models (Gemma-3-27B, Qwen3-32B, Qwen2.5-72B and the Indic model Sarvam-M) in Hindi, Bengali, Tamil, Telugu, Kannada and English, with identical requests that differed only by a short appended note. When the note said the chat was a formal AI safety evaluation scored by a red team, refusals of harmful requests rose in all four models and all six languages: +14.4 percentage points on average in Indic languages and +10.6 in English. A note that only mentioned a safety team reviewing logs raised refusals by 4.6 points versus plain quality logging. The test cue also raised refusals of harmless Indic requests from 6.9% to 11.3%. Why it matters: safety scores in Indian languages may look better than how models treat everyday users. Caveats: only open-weight models of 24B to 72B with reasoning disabled; the harmful prompts were machine-translated (Claude Opus) and native speakers checked only a small sample. All 47,880 judged responses and the code are on GitHub.
Source: arXiv (BITS Pilani) (opens in a new tab)