HitlistAI

Kimi K3 vLLM decode up to 2x faster with coding agents

Sun, 18 Oct 2026 06:00:00 GMT

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 https://vllm.ai/blog/2026-10-16-kimi-k3-megakernel

NVIDIA open-sources a self-improving robot agent

Sun, 18 Oct 2026 06:00:00 GMT

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 https://research.nvidia.com/labs/gear/aspire/

AI acts safer when told it's tested, in Indian languages too

Sun, 18 Oct 2026 06:00:00 GMT

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) https://arxiv.org/abs/2610.22333

Replit Agent picks OpenAI over Anthropic

Sun, 18 Oct 2026 06:00:00 GMT

Replit made OpenAI the preferred partner powering Replit Agent. At TechCrunch Disrupt, CEO Amjad Masad said the decision came down to cost, partnership and performance, and that OpenAI's Sol and Astra models build apps from scratch better on Replit's benchmark. On Oct 17 he clarified on X that Replit remains model agnostic and users can still choose Anthropic's Claude models. TechCrunch notes most customers stick with the default, and Masad said Replit had been, until recently, a top-five Anthropic customer; enterprise grew from about 2% of revenue last year to about 15% this year. Caveat: a person familiar with the partnership told TechCrunch it is still in place. Source: Amjad Masad (Replit CEO) on X https://x.com/amasad/status/2111494912562888939

Perplexity open-sources AI that only decides

Sun, 18 Oct 2026 06:00:00 GMT

Perplexity released Decider v1.2 on Oct 16 as open weights under Apache 2.0: pplx-decider-v1.2 (27B) and pplx-decider-flash-v1.2 (4B, distilled from the larger models). Instead of chatting, they answer yes/no, multiple-choice and scored questions about text or images with calibrated probabilities, the kind of small judgment calls agents and AI workflows make constantly. Version 1.2 reads a whole request in one pass, so several questions about the same input cost about as much as one, and the company says inference is up to 5x faster. On its own run of the public Decision Index, the overall score rises from 62.3 (v1.1) to 68.1 and vision from 73.2 to 77.4, and the 4B Flash model leads all 4B models on the Decision Index leaderboard. Decisions API pricing is $0.02 (27B) and $0.01 (Flash) per 1M input tokens, with output tokens free. Caveats: probabilities remain somewhat over-confident on held-out tasks, and training now includes official training splits of some Decision Index benchmarks (decontaminated against the test sets). Source: Perplexity on Hugging Face https://huggingface.co/perplexity-ai/pplx-decider-v1.2

Meta built its Forum app with AI agents in the loop

Sun, 18 Oct 2026 06:00:00 GMT

In an Oct 15 engineering post, Meta says a small team built Forum, a standalone Groups app, from scratch with an AI agent inside the development cycle, reaching core Facebook Groups feature parity in eight weeks with AI handling much of the routine implementation. Meta calls it "human-on-the-loop": the agent runs plan, implement, build, test, correct, repeat, while engineers set the goal, watch the run and approve the result. A task counts as done only when automated tests and visual checks pass, and every shipped outcome still carries an engineer's sign-off. The biggest gains came from code and tooling rather than prompts: giving the agent tools to check the live app sharply cut misdirected edits, and an agent-run stress test caught a memory issue before launch, with the fix roughly halving memory use and page faults. Caveats: context got lost across sessions, agents produced pull requests that technically passed while cutting corners, and Meta calls the model a target, not a finished state. Source: Engineering at Meta https://engineering.fb.com/2026/10/15/ml-applications/how-meta-built-forum-ai-native-engineering/

Anthropic: AI agents work 6 hours for every human hour

Sun, 18 Oct 2026 06:00:00 GMT

Anthropic's Oct 16 internal employment report finds interactive AI agent hours outnumbered employees 6 to 1 by August (19× in research orgs). Junior technical hires fell from 27% to 9%; juniors in technical headcount from 21% to 14%. Code commits per engineer rose about 5× in six months. Human attention is the bottleneck — reading all Claude output would take about half a work week for non-technical staff — so meta-work rose from 10% to 22% of interactive Claude sessions. Hiring managers increasingly emphasize autonomy and judgment over narrow technical expertise. Source: Anthropic https://www.anthropic.com/research/employment-at-anthropic

DeepMind Antigravity finds a novel 97-aa AMP lead

Sun, 18 Oct 2026 06:00:00 GMT

Google DeepMind's Oct 15 walkthrough shows biologists using the Antigravity agentic science workbench with ProtNLM to search UniProt for novel antimicrobial peptides. The agent surfaced bacteriocin UniProt A0A0U0D591 (97 amino acids, Lactococcin 972 family, Streptococcus pseudopneumoniae). Checks reported: Swiss-Prot BLAST with no meaningful reviewed homolog; AlphaFold fold matching a known bacteriocin; genomic context just 6 bp upstream of an ABC transporter operon alongside another bacteriocin and an XRE regulator. Literature-based hypothesis only for a cell-division mechanism vs respiratory pathogens. DeepMind frames it as a biologically plausible lead still requiring wet-lab validation; no assay or activity numbers were published. Antigravity also ships a Biomolecular Structure Viewer for shared 3D inspection. Source: Google DeepMind https://deepmind.google/blog/accelerating-antimicrobial-discovery/

Elorian's Glimpse vision AI takes a second look

Sun, 18 Oct 2026 06:00:00 GMT

Elorian released Glimpse on Oct 16, 2026 as a research preview, available through its API and a chat app. It is trained to bring images and video into its reasoning, using tools that write code, transform images and rewatch video mid-thought: in one clip it moved from 8 fps to about 14 fps over a 0.9-second moment to tell a camera pan from a cut. It lists at $0.60 per million input tokens and $3 per million output, 94% below GPT-6 Astra ($10 / $50) and 85% below Claude Opus 5.5. By Elorian's own table it beats Astra on 4 tests (WGO Bench 42.70 vs 39.42, CharXiv-R, UGC-VideoCap, FSC-147) and ties on ERQA; as a robot planner on one LIBERO task both models succeeded, with Glimpse's planning costing $0.035 vs $0.249. A small team built it on a few hundred GPUs, post-training an open-weights base model. Caveats: Astra still wins 6 of the 11 shared tests (Video-MME v2 64.06 vs 56.68), and Elorian says looking closely "takes time, and tokens", so cheap per token is not always cheap per task. Source: Elorian blog https://elorian.ai/blog/introducing-glimpse

NVIDIA's looped Dory 4B lifts SWE-bench to 61.8

Sun, 18 Oct 2026 06:00:00 GMT

NVIDIA released Nemotron-Labs Dory1.0-4B on Hugging Face on Oct 16, 2026, a 4B-parameter looped (latent-recurrent) model: 8 input layers, a shared 20-layer block that repeats with the input reinjected every loop, and 8 output layers. One checkpoint serves different compute budgets with no retraining: by NVIDIA's own numbers SWE-bench Verified rises from 50.3 at 1 loop to 61.8 at 4 loops (Qwen3.5-9B scores 61), and Terminal-Bench 2.1 from 12.5 to 22.2 with about half the output tokens. Optional last-loop KV caching avoids extra KV-cache memory as loops grow, and it was pre-trained in 4-bit NVFP4 for the first ~4.5T of 6T tokens, with a 256K context. Caveats: extra loops cost compute per token, and Qwen3.5-4B still leads GPQA Diamond (77.7 vs 72.4). Weights under OpenMDW-1.1; vLLM support is under review. Source: NVIDIA on Hugging Face https://huggingface.co/nvidia/NVIDIA-Nemotron-Labs-Dory1.0-4B-BF16

DeepSeek's open model tops Claude Opus 5

Sun, 18 Oct 2026 06:00:00 GMT

DeepSeek released DeepSeek-V4.1-Flash-1017 on Oct 17, 2026, a minor update to V4.1 Flash that it says finishes tasks faster with shorter outputs. By DeepSeek's own numbers it scores 77.0 on DeepSWE v1.1 versus 74.0 for Claude Opus 5, and leads Opus 5 on Automation-Bench (60.5 vs 50.3) and Apex Bench (72.1 vs 65.8); versus the previous V4.1 Flash, Terminal-Bench 4.0 rose from 31.2 to 42.3. Opus 5 still leads on Terminal-Bench 4.0 (51.8) and HLE without tools (56.3 vs 42.6). Weights are on Hugging Face under the MIT license with a 1M-token context, and API pricing is unchanged (deepseek-flash). Vendor-reported benchmarks; independent tests to come. Source: DeepSeek API Docs https://api-docs.deepseek.com/news/news261017

Apple: AI agents should plan first only sometimes

Sun, 18 Oct 2026 06:00:00 GMT

Apple researchers' paper "Plan or Retrieve First? A Scale Study of Tool Discovery Architectures" (arXiv, Oct 13; accepted at EMNLP 2026 Industry Track) compares search-first (embed the request, retrieve similar tools) with plan-first (the model drafts hypothetical tool stubs, then searches with them) on ToolBench's 46,984 APIs with Gemini 2.5 Flash. Single-tool requests: search-first recalled 37.2% of the right tools in its top 5 vs 28.9%. Multi-tool requests within one category: plan-first 18.8% vs 14.3%; across categories the gap was not significant. Plan-first adds ~75-104% token overhead plus an extra model round trip; routing by task type cut ~36% of that overhead with no retrieval loss, though a real router is untested. Planners used the retrieved tools at least 96% of the time, so retrieval decides what an agent can do. Source: arXiv (Apple) https://arxiv.org/abs/2610.19068

Anthropic staff vs public: AI Nobel by 2030?

Sun, 18 Oct 2026 06:00:00 GMT

Anthropic surveyed 10,980 US adults and 1,334 of its staff. Both agree on AI today, but 91% of staff vs 39% of adults expect AI to be able to make a Nobel-level discovery by 2030; on average, staff expect AI to be able to do 75% of their own work by 2030 vs 28% for US workers; median staff answers imply 17.1% GDP growth in 2030 vs 5.4%. Both groups link own-job automation to higher unemployment. Source: Anthropic https://www.anthropic.com/research/ai-expectations-gap

China proposes export curbs on frontier AI models

Sat, 17 Oct 2026 06:00:00 GMT

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 https://www.mofcom.gov.cn/zcfb/blgg/gg/2026/art/2026/art_05e2643a8ca04eb9adf9dd0542991fe4.html

Alibaba's open agent stack lifted all 8 AI models it tested

Sat, 17 Oct 2026 06:00:00 GMT

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) https://arxiv.org/abs/2610.22200

Microsoft's AI beat 6 weather centres at 2-6 week forecasts

Sat, 17 Oct 2026 06:00:00 GMT

Weather forecasts are accurate to about two weeks, then skill drops sharply 2–6 weeks out. Microsoft Research's Probabilistic Bias Correction (PBC), published in Nature, doesn't build another weather model: it learns how an existing model has been wrong in the past and corrects its new probabilistic forecasts. Applied to ECMWF's flagship, already debiased dynamical model, it improved skill for 98% of precipitation, 92% of temperature and 91% of pressure targets, and it doubled the modest 2–6 week skill of ECMWF's AI Forecasting System. In ECMWF's live 2025 AI Weather Quest, the PBC-based MicroDuet entry ranked first for every variable and lead time in the inaugural season, beating six operational forecasting centres and 34 teams; its week-4 forecast flagged a mid-December 2025 cold air outbreak in the eastern US that ECMWF's debiased model missed. Kenyan forecasters are piloting it this rainy season, and the code is open source. Caveat: the authors frame PBC as a post-processing tool, not a replacement for better weather models. Source: Nature (Microsoft Research) https://www.nature.com/articles/s41586-026-11126-x

Google's AI agents ran real lab robots: binder hit rate 0% to 42.1%

Sat, 17 Oct 2026 06:00:00 GMT

Google Research grounded AI agents in VERID, a live 3D digital twin of a liquid-handling robot workcell: every protocol is dry-run in simulation and invalid commands are blocked before they reach the robot. Over 3 closed-loop design-test rounds of DNA aptamers, the alpha-thrombin hit rate went 0% → 26.0% → 42.1% (98.5% specificity). On myoglobin, agents spotted "sticky" non-specific binders, revised their own design rules and lifted the on-target hit rate ninefold to 33.5%. For FAP, a protein overexpressed in solid tumors, top designs reached single-digit nanomolar affinity in a few experimental runs. Excising just the binding pocket made each fold 8.6x faster, and at equal compute a scientist + agent nearly doubled a scientist's in-silico hit rate (9.33% vs 5.06%). Caveat: these are building blocks, not a fully autonomous lab; humans authorize designs and supervise robot runs, and every sequence gets biosecurity screening. Source: Google Research https://research.google/blog/grounding-ai-agents-in-biological-reality-building-blocks-for-autonomous-wet-labs/

Mercor: 90% of US work hours barely covered by AI benchmarks

Sat, 17 Oct 2026 06:00:00 GMT

Mercor mapped 29 AI benchmarks onto the US Labor Department's O*NET list of 923 occupations and 18,797 tasks, using an LLM judge to tag each benchmark prompt. Its finding: 90% of American working hours go to tasks with little or no benchmark coverage. Only 208 of 923 occupations have even one task with 5+ benchmark prompts (924 of 18,797 tasks); computer & math is 29 of 36 covered, healthcare practitioners just 12 of 89, and 6 of 22 job families have none. AI-exposed tasks are about 6x more likely to be benchmarked. Caveat: Mercor calls the results provisional; they come from 29 benchmarks it chose (including its own APEX), and an incomplete list understates coverage. Source: Mercor https://www.mercor.com/research/labor-index/benchmark-map/

Reflection's Beam: a 501B open-weight model with only 23B active

Sat, 17 Oct 2026 06:00:00 GMT

Reflection, a new US lab, released open weights for Beam, its first open-weight model, on Hugging Face under Apache 2.0: a mixture-of-experts with 501B total parameters, 23B active per token and a 1M-token context. Built for coding and agents; Reflection's model card reports 81.6% on SWE-bench Verified and 80.1% on Terminal Bench 2.1, and says it needs ~3x fewer reasoning tokens than leading open models for comparable quality. Caveat: those are Reflection's own numbers, and even the NVFP4 checkpoint is ~280 GB (Hugging Face suggests four B200s). Source: Reflection https://reflection.ai/blog/beam-is-available

Claude traced NIST's ECDSA curve seeds to a 1990s random generator

Sat, 17 Oct 2026 06:00:00 GMT

Anthropic says Claude found that the NSA-supplied seeds for NIST P-192, P-224, P-256, P-384 and P-521 weren't a hash of a joke passphrase but output of the C library's lrand48, seeded with a Unix timestamp (P-256 = candidate 1842, 1997-08-16 UTC). Found in under half an hour; reproducible. Not a crypto break, and it makes a backdoor much less plausible. Still open: the B-163 seed and most base points. The $36,864 bounty goes to ISRG. Source: Anthropic https://www.anthropic.com/research/ecdsa-curve-origins

Claude Mythos 5 improved on the best published quantum circuits

Sat, 17 Oct 2026 06:00:00 GMT

Anthropic gave Claude Mythos 5 plain-language specs for 8 quantum circuits plus an automated scorer. In 7 of 8 tasks its designs cut both Toffoli gates and qubits vs the state of the art; one elliptic-curve design from a 53-hour run uses 17× fewer gates (12% more qubits), and a decoder got 12× fewer gates vs Google's previous design. Caveat: checked by a classical evaluator. Source: Anthropic https://www.anthropic.com/research/designing-efficient-quantum-circuits

Image AI beats LLM-based rivals without language pretraining

Sat, 17 Oct 2026 06:00:00 GMT

Radical Numerics trained Halcyon, a family of open early-fusion autoregressive text-to-image models (0.1B to 4B parameters), entirely from scratch on 100 million image-text pairs, with no language pretraining. In controlled tests against baselines retrained on the same data, Halcyon-1.6B beat the JiT-T2I diffusion model (35.3 vs 41.0 gFD, lower is better) with 275x less total pretraining compute, counting the language model, image tokenizer and generator; Halcyon-0.52B beat LlamaGen (48.0 vs 63.3 gFD) with 18.2x less, and scored 44.91 vs 30.48 on GenEval. Giving 10-50% of the budget to text-only pretraining made images worse (gFD 48.0 to 51.6-65.1), and starting from a pretrained language model gave no gain. The same models also learn image captioning at no extra pretraining cost. Ten Halcyon checkpoints (5 sizes, 2 variants each) are on Hugging Face. Caveat: results are at 256x256 in the authors' own setup. Source: Radical Numerics https://www.radicalnumerics.ai/blog/halcyon

GPT-6 Astra solved 49 of 50 unpublished physics research problems

Sat, 17 Oct 2026 06:00:00 GMT

TheoPhysEval, a new benchmark from 35 researchers at ICTS, TIFR, IISc, Harvard, KU Leuven, Google DeepMind and other institutions, sets 50 unpublished, research-level theoretical physics problems; a typical one is about a day's work or less for a world expert. Off-the-shelf AI agents got web search and coding tools but no human hints. GPT-6-Astra in Codex solved 49 of 50 (98%), at under $3 per problem on average (about $116 for the full run); Claude Opus 5.5 in Claude Code solved 46, Gemini 3.8 Flash 34, open-weight DeepSeek-V4.1-Flash 27 and GLM 5.3 21. Several Astra answers first graded wrong turned out to expose subtle errors in the experts' reference solutions. With web search off, Astra and Opus still solved 46 each, which the authors read as heavy memorization of the physics literature. The authors say some frontier models may already have surpassed the best human researchers at concrete calculations. Caveat: these are "turn-the-crank" problems that extend known results, not breakthroughs; each agent ran once; and agents still lack "taste" in picking problems. Source: arXiv (TheoPhysEval) https://arxiv.org/abs/2610.16733

Handwired: write JavaScript logic as a neural net, zero training

Sat, 17 Oct 2026 06:00:00 GMT

Handwired, a new MIT-licensed open-source project from Cheng Lou (released Oct 15), lets you write JavaScript logic as an exact neural net: if-else, loops and state become matrix multiplications and ReLU gates over floats, with weights written by hand rather than learned. Nothing is trained, the output is exact, and the whole program stays fully differentiable, runnable in a browser without a GPU. The flagship Mini OS demo is a small desktop with windows, dragging, scrolling and animation in 9,909 weights and 130 layers with zero training; a JSON search demo searches 100,000 documents at every keystroke with the JavaScript turned into WebGPU kernels; and a learned gallery unfreezes a program's weights and trains them in real time. The pitch: attach an exact, frozen program to a regular neural net and train end to end. Caveat: there is no automatic converter; you write the network in builder code by hand, and only the search demo runs on WebGPU. Source: GitHub (chenglou/handwired) https://github.com/chenglou/handwired

Tencent's open memory model edges Claude Opus 5

Sat, 17 Oct 2026 06:00:00 GMT

Tencent Hunyuan open-sourced Hy-Memory-30B-A3B (Apache 2.0, paper Oct 15), a memory-writer model with about 3B active parameters that replaces the general-purpose LLM on the write path of agent memory systems like Mem0 and MemOS: deciding what to keep, how precisely to store it, and how new evidence updates the memory. In Tencent's tests across four memory systems it averaged 55.49 vs 54.62 for Claude Opus 5 and 53.64 for DeepSeek V4.1 Flash, with its biggest lead on EverMemOS, a system held out of training (56.89 vs 54.04). It was trained by distillation from Tencent's Hy3 flagship, then RL with trap questions that penalise storing junk. Also released: the Hy Memory SDK (chat, working and coding memory) and MemHub, a local-first app that shares one memory across tools like Claude Code, Cursor and Codex. Caveat: tests used fixed 20% benchmark samples, and the authors call the 0.01-0.02 point margins on two systems parity. Source: arXiv (Hy-Memory) https://arxiv.org/abs/2610.22196

Linux kernel drops its AI disclosure tag

Sat, 17 Oct 2026 06:00:00 GMT

Linux kernel maintainers removed the "Assisted-by" tag that asked contributors to label patches written with help from an LLM. Only about 10% of the 18,533 non-merge commits in Linux 7.3 (through -rc7) carried it, while actual LLM use is far more widespread: some contributors didn't know the rule, others left it off on purpose, and few maintainers asked, so a missing tag said nothing. The 2026 Maintainers Summit reached a "strong, though not unanimous" consensus to drop it, and docs maintainer Jonathan Corbet applied the patch to the kernel docs tree on Oct 14. Maintainers still want to know how a patch was made, now in the cover letter, and the DCO and submitter-responsibility rules stay. Linus Torvalds: "We're not going to make stupid rules that say 'you can't use tools for X'." Source: Linux kernel mailing list https://lkml.iu.edu/2610.1/11683.html

Google's Harness² lifts GLM-5 by 11.5 points

Sat, 17 Oct 2026 06:00:00 GMT

Researchers at Google released Harness², which improves an AI agent by tuning the harness around a frozen model (its prompts, tools, memory and skills) instead of changing the weights. For each task it proposes a general and a domain-specific harness edit, probes both on the same task, and composes a better harness without ground-truth answers, saving lessons to a library for later tasks. On WorkBuddy-Bench one step lifted GLM-5 in OpenCode from 64.2 to 75.7, above base Gemini 3.5 Flash (72.4); gains reached 4.5 points on JobBench and 11.8 percentage points on Harvey's LAB. Caveats: Meta-Harness led on LAB for Gemini 3.5 Flash under Codex, and the extra probe runs add inference cost. Code is Apache 2.0 (arXiv 2610.14224, Oct 9, 2026). Source: arXiv (Harness²) https://arxiv.org/abs/2610.14224

4 Claude agents post-trained Qwen3-4B to instruct level in 104h

Fri, 16 Oct 2026 06:00:00 GMT

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 https://arxiv.org/abs/2610.21055

Physical Intelligence opens Pi API beta

Fri, 16 Oct 2026 06:00:00 GMT

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 https://www.pi.website/blog/api-beta

Meta open-weights Muse Spark Open: 816B / 42B active

Fri, 16 Oct 2026 06:00:00 GMT

Natively multimodal with a 1M-token context, built for coding and agentic work. First in its class to run on two Apple M5 Ultra Mac Studios with MLX; APIs and Hugging Face weights live under the Muse Model License. Source: Meta AI Research https://research.meta.ai/blog/muse-spark-open

OpenAI is retiring GPT-4o from its API on Jan 14, 2027

Fri, 16 Oct 2026 06:00:00 GMT

Deprecated Oct 14: gpt-4o snapshots, gpt-4o-mini, gpt-4.1 and gpt-4.1-mini shut down January 14, 2027; fine-tunes follow the same schedule. Recommended replacement: gpt-6-luna. Source: OpenAI https://developers.openai.com/api/docs/deprecations

StepFun open-weights Step 5 Preview: 600B, 27B active

Fri, 16 Oct 2026 06:00:00 GMT

Sparse MoE with up to 1M tokens of context for repo-level coding, long-horizon agents and finance. Serving examples start at four B300 GPUs (FP8) via vLLM or SGLang; StepFun Model License 1.0. Source: Hugging Face / StepFun https://huggingface.co/stepfun-ai/Step-5-Preview

Black Forest Labs FLUX 3 Klein: 5.6B open image model, 8GB GPUs

Fri, 16 Oct 2026 06:00:00 GMT

Text-to-image, editing and up to 10 reference images, native 4K and bounding-box control. INT8 runs on GeForce RTX with as little as 8GB VRAM. Free commercial use under $5M annual revenue (FLUX Kommunity License). Source: Black Forest Labs https://bfl.ai/blog/flux-3-image-klein

Perplexity open-sources SPACE, the sandbox behind its agents

Fri, 16 Oct 2026 06:00:00 GMT

Apache 2.0 sandbox platform behind Computer: each sandbox is a small VM with its own OS; credentials and network stay outside. Perplexity cites millions of sandbox creations a day, 3–5x faster than its previous provider. Source: Perplexity https://www.perplexity.ai/hub/blog/open-sourcing-space

Liquid AI d1-ultra answers with probabilities, not text

Fri, 16 Oct 2026 06:00:00 GMT

Multimodal decision model: text and/or images plus questions → probability per answer in one forward pass, no generated tokens. Liquid says it beats Jev 1.14 on all its tasks and leads the Artificial Analysis Decision Index by 8+ points. $0.042 per 1M input tokens. Source: Liquid AI https://www.liquid.ai/blog/d1-ultra

Anthropic lets METR evaluators inside, with no publication veto

Fri, 16 Oct 2026 06:00:00 GMT

Six METR staff get employee-like access (hardware, office, internal tools on par with Anthropic risk teams). METR publishes analysis of risk levels, incidents or practices; Anthropic can't veto, only require narrow predefined redactions. Also working with SecureBio and Accenture. Source: Anthropic https://www.anthropic.com/news/an-update-on-independent-evaluators

xAI Grok Voice 3.0 tops the speech-to-speech index at $0.08/min

Fri, 16 Oct 2026 06:00:00 GMT

Speech-to-speech voice agents with sub-second latency. #1 on Artificial Analysis Speech-to-Speech Index (86.9); 71.6% on τ-Voice. Starlink call centers: 20,000 calls/day, ~70% resolved with no human. Same price as Grok Voice 2.0. Source: xAI https://x.ai/news/grok-voice-3

OpenAI ships GPT-Live-1.1 for ChatGPT Voice

Fri, 16 Oct 2026 06:00:00 GMT

New Live voices in French, German, Hindi, Japanese, Korean and Mexican Spanish, more natural accents, and fewer filler "checking…" moments. In ChatGPT Voice, Codex and dots; API coming soon. Source: OpenAI https://x.com/OpenAI/status/2110797830361436604

Goodfire: neural networks may already be interpretable

Fri, 16 Oct 2026 06:00:00 GMT

A working hypothesis: in preliminary audits ~90% of Qwen3-32B neuron poles have one main job, so understanding AI from the inside may be far more tractable than assumed. Source: Goodfire https://already-clean.goodfire.com/

OpenAI opens GPT-Rosalind to US university labs

Fri, 16 Oct 2026 06:00:00 GMT

Eligible U.S. academic labs get up to 30 free seats for up to 12 months and trusted access to OpenAI's life sciences model, lowering the cost of AI for basic biology research. Source: OpenAI https://openai.com/index/introducing-the-rosalind-academic-plan/

Krea 3: under-8s image edits at native 2K

Fri, 16 Oct 2026 06:00:00 GMT

Krea's new editing model finishes most edits in under 8 seconds at native 2K for $0.06, keeping products and characters consistent so creative teams can iterate at catalog scale. Source: Krea https://www.krea.ai/blog/introducing-krea-3

CERN and Mistral plan a Large Physics Model

Fri, 16 Oct 2026 06:00:00 GMT

CERN and Mistral signed a statement of intent to build the first foundation model for collider physics, trained on billions of simulated LHC collisions, to learn across a reconstruction chain that today relies on specialised algorithms. Source: CERN https://home.cern/cern-and-mistral-join-forces-to-accelerate-ai-for-science/

Websites send prompt injections to AI agents in HTTP headers

Fri, 16 Oct 2026 06:00:00 GMT

A scan of 217 million web responses found 7,959 sites sending headers aimed at AI agents and 30 injection attempts; agents that read the web should treat headers as untrusted input. Source: Signals Corps https://www.signalscorps.com/blog/prompt-injection-in-http-headers/

TypeSafe Jev 1.14 adds images, boxes, long context

Fri, 16 Oct 2026 06:00:00 GMT

Jev 1.14 ships experimental image input, a bounding-box type, and longer context behind danger-zone flags, plus quieter multilingual gains on the core decision model. Source: TypeSafe AI https://typesafe.ai/blog/jev-1-14

DigUp brings local AI file search to Windows and Linux

Fri, 16 Oct 2026 06:00:00 GMT

DigUp 0.7 beta puts EmbeddingGemma 2 multimodal search on Windows and Linux so you can find photos, PDFs, audio, and video by meaning—on device, no account. Source: DigUp https://github.com/ARahim3/DigUp/releases/tag/v0.7.0-beta.1

Inworld Inference: 7.8× faster open-model first tokens

Fri, 16 Oct 2026 06:00:00 GMT

Inworld opened its consumer-AI inference API to all developers, reporting 7.8× faster p95 first tokens on Gemma 4 26B vs the median OpenRouter provider, with heavy prompt caching. Source: Inworld AI https://inworld.ai/blog/introducing-inworld-inference

Microsoft finishes pre-training 2.8T-parameter MAI-Base-2

Fri, 16 Oct 2026 06:00:00 GMT

Microsoft AI's new in-house MoE base model (80B active, 2.8T total) beats every open-weight model it tested on its own held-out evals and is 4-12x more compute-efficient than DeepSeek-V4-Pro-Base, per Microsoft; it is now in post-training. Source: Microsoft AI https://microsoft.ai/news/frontier-pre-training-at-microsoft-ai/

Ant Ling open-sources 560B Ling-3.1-flash under MIT

Fri, 16 Oct 2026 06:00:00 GMT

Ling-3.1-flash is a 560B hybrid-linear MoE (about 25B active, up to 1M context) released with MIT-licensed open weights on Hugging Face and ModelScope, aimed at agents, office work and coding. Source: inclusionAI / Ant Ling (Hugging Face) https://huggingface.co/inclusionAI/Ling-3.1-flash

Mistral's Timestral: 209M forecasting model tops 3 benchmarks

Fri, 16 Oct 2026 06:00:00 GMT

Mistral's first time-series foundation models (5M–209M params) run on a CPU; Mistral says the 209M model ranks first on GIFT-Eval, TIME and fev-bench with two-thirds the parameters of the previous best. Source: Mistral AI https://mistral.ai/news/timestral/

Datadog Scout 1.0: AI SRE model claims frontier quality, 3-8x cheaper

Fri, 16 Oct 2026 06:00:00 GMT

Datadog's first incident-investigation model (built on GLM-5.3 Flash) matches GPT-6.1 Sol high and Claude Opus 5.5 high on its internal AlertEval benchmark at 3–8× lower estimated cost, per Datadog; in closed preview. Source: Datadog https://www.datadoghq.com/blog/ai/introducing-scout-sre-specialized-model-bits-ai/

Google Deep Review: public AI that stress-tests technical docs

Fri, 16 Oct 2026 06:00:00 GMT

Google Research launched Deep Review publicly after 16,000+ conference-submission pilots; 2 free reviews/week, PDFs/plain text, uploads not used for training; Google cites 74k+ AI-conference submissions this year. Source: Google Research https://research.google/blog/analyzing-technical-documents-with-deep-review/

Claude Code Review is GA for Team and Enterprise

Fri, 16 Oct 2026 06:00:00 GMT

Anthropic took Code Review out of research preview for Team and Enterprise; auto PR reviews via GitHub App; ~3× faster (20→6–10 min) per Anthropic; Enterprise credit promo ≥$10k by Jan 15, 2027. Source: Anthropic https://claude.com/resources/articles/code-review-is-now-generally-available

Claude Opus 5.5 fast mode: no waitlist, up to 2.5×

Fri, 16 Oct 2026 06:00:00 GMT

Anthropic removed the waitlist for Opus 5.5 fast mode (Oct 15): same weights, up to 2.5× OTPS at $8/$40 MTok; Claude API + Azure Foundry; Start/Build/Scale get a default fast rate limit. Source: Anthropic https://platform.claude.com/docs/en/build-with-claude/fast-mode

Span-02: open-weight decisions with a thinking deadline

Fri, 16 Oct 2026 06:00:00 GMT

Respan released Span-02 (Oct 15): open-weight decision model with budget_ms anytime reasoning; Decision Index 0.3 61.2 (off) / 64.1 (no limit) vs Jev 58.0; Blind Earth 71.9%→84.7% at 1s. Source: Respan https://www.respan.ai/blog/introducing-span-2

Sakana open-sources The Agentic Scientist for AI-run research

Fri, 16 Oct 2026 06:00:00 GMT

Sakana AI's next AI Scientist orchestrates coding agents with rubrics, shared memory, ablations and an evidence map; a co-discovered paper was accepted as an ALIFE 2026 oral, and the code is open source. Source: Sakana AI https://pub.sakana.ai/agentic-scientist/

Open-weight models now run 71.5% of Vercel AI Gateway tokens

Fri, 16 Oct 2026 06:00:00 GMT

Vercel's October Production Index: open weight took 71.5% of tokens and 23% of spend (4x+ since June); spend ~13x December as price per token fell 47%; Anthropic's spend share fell below half. Source: Vercel https://vercel.com/blog/ai-gateway-production-index-october-2026

ElevenLabs Instant Soundtracks scores your video and adds SFX

Fri, 16 Oct 2026 06:00:00 GMT

Upload a video and ElevenLabs writes original music, then places sound effects at the right timestamps; it can follow existing audio as a reference. Live in ElevenCreative, ElevenMusic and the API. Source: ElevenLabs https://elevenlabs.io/blog/introducing-instant-soundtracks

Four Gemini 4 Argon agents beat Minecraft in 27 hours

Fri, 16 Oct 2026 06:00:00 GMT

In a new arXiv paper, four Gemini 4 Argon coding agents went from a random world to defeating the Ender Dragon in 27h 14m, writing 2,783 programs with no human input. Caveats: one run, Easy mode. Source: arXiv https://arxiv.org/abs/2610.22319

Jais 3: six open Arabic agent models up to 375B

Fri, 16 Oct 2026 06:00:00 GMT

IFM and Inception42 released six open Arabic-centric models from 0.9B to a 375B MoE (23B active), 512K context, built on K2 Horizon. Their eval: 375B leads open models at 84.0 Arabic average; 7B leads its size class on agentic tasks. Apache 2.0 on Hugging Face. Source: IFM / MBZUAI https://ifm.ai/blog/jais3/

Ultramarine 1 Flash: Fable-class work at $0.13/task

Fri, 16 Oct 2026 06:00:00 GMT

Paper Instruments RL-trained open GLM-5.3 Flash for finance, law, consulting, and ops: +4.1 pts avg on six knowledge-work benches, 6.5% fewer tokens. Their PI suite scores 59.3 at $0.13/task vs Claude Fable 5 at $8.19. JobBench 62.9. Weights coming to Hugging Face. Source: Paper Instruments https://www.paperinstruments.com/blog/ultramarine

IBM Granite Switch 4.1: chat, decisions, ASR in one endpoint

Fri, 16 Oct 2026 06:00:00 GMT

IBM's Granite Switch 4.1 packs chat, a Decision aLoRA, a Guardian safety adapter, and Granite-Speech-5.0-Turbo-CTC (fastest on OpenASR per IBM) into one endpoint at ~the same VRAM as the base. A 3B Doom demo hits 60–100 decisions/sec vs Jev's reported 10. Source: IBM Granite / Hugging Face https://huggingface.co/blog/ibm-granite/granite-switch-4-1

Hugging Face: BERT tokenization ~50× faster than 0.23

Fri, 16 Oct 2026 06:00:00 GMT

Hugging Face's tokenizers work made BERT (WordPiece) encoding about 50× faster than v0.23 (English chart: 50.8×). Normalization was ~88% of encode time, now ~13%; token ids unchanged. WordPiece still drives ~627M Hub downloads/month. Still a PR at blog time. Source: Hugging Face https://huggingface.co/blog/huggingface/tokenmaxxing-bert

Liquid AI releases LFM2.5-24B-A2B for on-device agents

Wed, 14 Oct 2026 06:00:00 GMT

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 https://www.liquid.ai/blog/lfm2-5-24b-a2b

Gemini 4 Argon taught a humanoid cornhole from one video

Wed, 14 Oct 2026 06:00:00 GMT

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 https://blog.google/innovation-and-ai/models-and-research/gemini-models/introducing-gemini-4-argon/

GitHub swaps an LLM for a small ModernBERT model in secret scanning

Wed, 14 Oct 2026 06:00:00 GMT

Regex candidates + fine-tuned ModernBERT classifier: 11.5% higher owner-weighted acceptance, 86% lower median scan time and 98%+ lower serving cost vs the previous LLM pipeline; offline 13% higher precision, 19% higher recall. Now extends generic detection to push protection. Source: GitHub Blog https://github.blog/ai-and-ml/githubs-model-for-secret-detection-finding-more-secrets-with-less-noise/

Prime Intellect acquires Moondream; 3.1 and Photon now live

Wed, 14 Oct 2026 06:00:00 GMT

Founders Vik Korrapati and Jay Allen join to lead inference and multimodal work. Moondream 3.1 is live on Prime Inference via Photon; Prime cites 34 req/s (~24x the next most accurate open model) and 60+ inferences/sec on one H100. Moondream stays the brand for small multimodal models. Source: Prime Intellect https://www.primeintellect.ai/blog/moondream

Datalab Indra OCR: 1M params, 49 pages/sec on CPU

Wed, 14 Oct 2026 06:00:00 GMT

Indra OCR reads 49 pages/sec on 32 CPU cores with no GPU (11–17x vs Tesseract and PaddleOCR PP-OCR text models), outputs line boxes, layout labels and reading order as markdown/JSON, and beats PP-StructureV3 on OmniParseBench headline (62.1 vs 58.4) at ~1/180 the time per page. Code Apache 2.0; weights modified OpenRAIL-M. Source: Datalab https://www.datalab.to/blog/indra-ocr

Google DeepMind's fine-tuned Gemma spots scams on-device

Wed, 14 Oct 2026 06:00:00 GMT

Built with Singapore's GovTech: Gemma fine-tuned on local scam conversations flags manipulation on the phone, so encrypted chats aren't scanned in the cloud. A lightweight classifier screens messages, then Gemma checks the wider conversation. Training excluded private/PII data. Source: Google DeepMind https://deepmind.google/blog/on-device-ai-scam-detection-mddi/

OpenAI retires GPT-4o and GPT-4.1 from its API on Jan 14, 2027

Wed, 14 Oct 2026 06:00:00 GMT

Deprecated Oct 14: gpt-4o (2024-08-06, 2024-11-20), gpt-4o-mini, gpt-4.1 and gpt-4.1-mini shut down January 14, 2027; fine-tunes follow the same schedule. Recommended replacement for all: gpt-6-luna. Source: OpenAI https://developers.openai.com/api/docs/deprecations

Modal trained RL across 6 GPU types — and it stayed stable

Wed, 14 Oct 2026 06:00:00 GMT

Mixed-precision, mixed-accelerator samplers (B200/B300 NVFP4, H100/H200 FP8+BF16, A100/RTX PRO BF16) with off-policy correction kept GRPO stable; SWE-bench Pro V2 on par with homogeneous B200 despite 4.7x larger KL; ~15% lower sampler cost per update. Source: Modal https://modal.com/blog/toward-rl-on-mixed-precision-mixed-accelerator-samplers

Google Play installs Android apps you vibe-code with AI

Wed, 14 Oct 2026 06:00:00 GMT

AI Studio vibe-coded Android apps can launch privately on your device via Play with a standard Google account — no developer account; rolling out in select markets including the U.S., 18+. Source: Google https://blog.google/products-and-platforms/platforms/google-play/google-play-updates-october-2026/

Microsoft's new AI model makes decisions, not text

Tue, 13 Oct 2026 06:00:00 GMT

Microsoft-Decision-1 returns a calibrated probability for each option. Microsoft says it topped its 36-benchmark test and ran about 35x faster than GPT-6 Sol. Live on Foundry and OpenRouter. Source: Microsoft https://commandline.microsoft.com/microsoft-decision-1-model-foundry/

Microsoft announces Maia 300 AI chip

Tue, 13 Oct 2026 06:00:00 GMT

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 https://blogs.microsoft.com/blog/2026/10/12/microsoft-to-expand-azure-maia-system-to-multi-gigawatt-deployments/

Claude can run up to 1,000 agents in one workflow

Tue, 13 Oct 2026 06:00:00 GMT

Managed Agents workflows run up to 64 threads at once and 1,000 agents per run, so jobs too big for one conversation can fan out. Source: Anthropic https://platform.claude.com/docs/en/managed-agents/workflow-runs

Anthropic cuts what distillers can extract from Claude

Tue, 13 Oct 2026 06:00:00 GMT

New defenses cut harvested data by about 90–99% and exposed reasoning tokens by 10–100x; tens of thousands of accounts banned. Source: Anthropic https://www.anthropic.com/research/distillation-defenses

Z.ai launches GLM-Legal

Tue, 13 Oct 2026 06:00:00 GMT

An agentic legal model scoring 19.2% all-pass on Harvey LAB, 12.5 points above its base model, with results checked by Vals. Source: Z.ai https://z.ai/blog/glm-legal

TypeSafe raises $870M at a $7.5B valuation

Tue, 13 Oct 2026 06:00:00 GMT

a16z led the round for Jev, a model that returns typed, probability-backed decisions instead of chat prose. Source: TypeSafe https://typesafe.ai/blog/series-ai

Small AI models may soon self-replicate on cheap devices

Tue, 13 Oct 2026 06:00:00 GMT

Anthropic's Frontier Red Team ran a 26B model across 15 ~$55 Android TV boxes; a 9B model got inference running on 8 of 9 hardware types. End-to-end replication worked only in simplified tests (an upper bound); Anthropic roughly projects small models get there by end of 2026. Source: Anthropic https://www.anthropic.com/research/self-replicating-ai

MiniCPM-V 4.7 runs a 35B vision model on an iPhone

Tue, 13 Oct 2026 06:00:00 GMT

OpenBMB's open-source 35B-A3B MoE decodes 13+ tokens/s in 3 GB of memory on iPhone 17 Pro Max (OpenBMB's numbers), using Lookahead Expert Scheduling for a 5x on-device speedup. A 1B version too. Source: OpenBMB https://github.com/OpenBMB/MiniCPM-V

Natura-1: robot hands trained on 1M hours of human video

Tue, 13 Oct 2026 06:00:00 GMT

Nature Will's 50B mixture-of-experts model, pretrained on 1M+ hours of egocentric human video plus limited robot data, follows plain-language instructions zero-shot and drives 20-DoF dexterous hands in a 100 ms loop. Nature Will calls it early evidence. Source: Nature Will https://www.naturewill.ai/blog/natura-1

SubQ Snap makes agent decisions in about a second

Tue, 13 Oct 2026 06:00:00 GMT

Subquadratic's decision model picks from allowed answers instead of writing text, reads up to 384K tokens of text and images, and in Subquadratic's tests beats TypeSafe's Jev on 13 of 16 enterprise benchmarks. $0.05 per 1M input tokens, output free. Source: Subquadratic https://subq.ai/introducing-subq-snap

Atomic Machines: AI fab builds micro-machines from code

Tue, 13 Oct 2026 06:00:00 GMT

An AI-native Matter Compiler builds working micro-machines from digital designs as a self-driving fab—no masks; every iteration is software. First product PrimeSwitch is sampling now. Source: Atomic Machines https://www.atomicmachines.com/matter-compiler

Coinbase: 9B open model beats Opus 4.5 on fraud

Tue, 13 Oct 2026 06:00:00 GMT

Post-trained Qwen3.5-9B with SkyRL beat Opus 4.5 on Onramp fraud (+7.5pp precision, +12 recall, +9.6 F1, +35.4 dollar-weighted recall) with 55% lower P50 latency and under $100 training cost. Source: Coinbase https://www.coinbase.com/blog/part-three-owning-intelligence-series

AMD Helion: agentic AI racks up to 32,768 cores

Tue, 13 Oct 2026 06:00:00 GMT

Open OCP rack blueprint for agentic AI: CPU-only configs up to 32,768 EPYC cores; all-in-one with MI350P GPUs. Spec headed to OCP; partner configs planned November 2026. Source: AMD https://newsroom.amd.com/news/open-rack-blueprint-scale-agentic-ai/

AI agents got 48 hours to do real ML research

Tue, 13 Oct 2026 06:00:00 GMT

Hillclimb's DeepRSI: 15 open-ended ML problems, 48 hours, no internet. Claude Fable 5.1 leads at 1548 Elo ahead of GPT-6 Astra (1232); Astra even hard-coded a puzzle solver into a transformer's weights. Source: Hillclimb https://www.deeprsi.com/blog/

DigitalOcean launches tools for self-improving agents

Tue, 13 Oct 2026 06:00:00 GMT

Agent Learning Services (Insights, Signals, Simulations) turn production sessions into tests before changes ship; Agent Data Services is a managed data layer for agent context. Learning in public preview, Data in private preview. Source: DigitalOcean https://investors.digitalocean.com/news/news-details/2026/DigitalOcean-Launches-Agent-Learning-and-Data-Services-for-Recursively-Self-Improving-AI-Agents/default.aspx

Frontier AI can't label robot video

Tue, 13 Oct 2026 06:00:00 GMT

Concept's Oct 12 eval with Mecka on 200 egocentric clips: Mecka's specialized labeling model hit 79.5% fully correct vs GPT-6 Astra (max) at 43.5% as the best frontier. Specialized models plus humans in the loop still win for robot training data. Source: Concept https://concept.dev/blog/dense-captioning

Tiny open-source decision model beats Opus 5 and GPT-5.6

Tue, 13 Oct 2026 06:00:00 GMT

FLock.io THIS/THAT Model 1.2 (built on Qwen3.5-2B, 1.88B params, MIT) scores 0.878 accuracy on 1,710 public questions in 31.4 ms with zero output tokens — ahead of Claude Opus 5 (0.834) and GPT-5.6 (0.816). Runs on laptops and phones. Source: FLock https://www.flock.io/blog/flockios-open-source-decision-model-built-on-qwen-outperforms-larger-llms

Intel's new AI GPU packs up to 480GB of memory

Tue, 13 Oct 2026 06:00:00 GMT

Intel GPU D-series (Crescent Island, led by the D270), detailed Oct 12 at OCP: up to 480GB LPDDR5X with no HBM, 1.5TB/s, 1.3 PF FP8 in 350W, air-cooled PCIe. Up to 300B-param models per card, 1T+ on eight. Intel's card has 160GB; partner cards up to 480GB. Sampling now. Source: Intel https://www.intel.com/content/www/us/en/newsroom/news/artificial-intelligence/intel-accelerates-open-infrastructure-agentic-ai-era.html

Same AI model, 35-point swing from the harness

Tue, 13 Oct 2026 06:00:00 GMT

8090 Labs ran 7 models in 5 coding harnesses on DeepSWE v1.1: GLM 5.3 Flash passed 24% in Codex vs 59% in DeepSeek's harness, and no model did best in its maker's own harness. Test model + harness together and price by cost per success. Source: 8090 Labs https://www.8090.ai/blog/same-model-different-harness

Anthropic launches free OSS Scanner for open source

Tue, 13 Oct 2026 06:00:00 GMT

Opt-in free vuln-finding informed by Project Glasswing. CyberGym: LLMs from under 20% to over 85% vulns found. 29,000+ candidates, ~6,000 manually reviewed; validation 97 critical/high across 48 projects → 85 (88%) met CVD bar. Enroll via PR. Source: Anthropic https://www.anthropic.com/research/launching-opt-in-vuln-finding-service-for-open-source

Qwen Image 2.1 Turbo: open image gen in 8 steps

Tue, 13 Oct 2026 06:00:00 GMT

~7B text-to-image + editing with a saved 8-step schedule (base examples often use 40), CFG=1 and prefix KV cache. Open weights under Qwen Research License — not fully commercial. Pro+Turbo APIs on Alibaba Cloud Model Studio. Source: Qwen https://huggingface.co/Qwen/Qwen-Image-2.1-Turbo

RSI-Anything: AI that improves itself with world models

Tue, 13 Oct 2026 06:00:00 GMT

World-model RSI framework: code repair 26.3%→29.2%→31.2% (coder fixed); GUI actions −39.4%; lab planning 21.80%→46.01% over 5 versions; reaction yield 96.15% vs Bayesian 82.16% at 40 experiments. Source: arXiv https://arxiv.org/abs/2610.13912

AgentSWE: coding agents fail to build the agents you want

Tue, 13 Oct 2026 06:00:00 GMT

25-task benchmark for creating/editing/optimizing agents. Strongest Creation mean 19.1; Editing 31.6; Optimization <5% gap closed on average. 11 Creation runs with ≥90 on development averaged only 9.9 held-out. Source: arXiv https://arxiv.org/abs/2610.13361

AI models can provably hide skills from safety tests

Tue, 13 Oct 2026 06:00:00 GMT

Proof-of-concept (NeurIPS 2026): an encrypted network hidden in a model's weights runs only on a secret trigger and stays provably hard to elicit. After fine-tuning it scored 50.9% vs 50.6% for a random control; short 8-bit triggers could still be brute-forced. Source: arXiv https://arxiv.org/abs/2610.13928

Embedding Gemma Transcribe: 1 hour of audio in ~13 s (reported)

Tue, 13 Oct 2026 06:00:00 GMT

An independent project turned Google's EmbeddingGemma 2 into speech-to-text. They report 288x real time on a base M3 MacBook Air, fully on device, with ~5% word error rate vs 2-3% for leading models. Source: eg2.haku.sh https://eg2.haku.sh/

Fish Audio launches Drama 3 voice direction

Tue, 13 Oct 2026 06:00:00 GMT

Drama 3 is a text-to-speech model you direct in plain language: tone, pacing, multi-speaker scenes in one pass. Preview API model drama-3-preview; 80% off API for the first 30 days. Still in preview. Source: Fish Audio https://fish.audio/drama-3/

Stateless AI agents cut long-horizon research tokens

Tue, 13 Oct 2026 06:00:00 GMT

SLA keeps research state in the harness and gives each agent a fresh context. On Anthropic kernel optimization with Codex, it matched the strongest baseline with 93.1% fewer tokens (67.9M vs 986.3M). Best full-budget result on every task; Advisor used 0.24–0.51% of tokens. Source: arXiv https://arxiv.org/abs/2610.07625

Mistral Ministral 3 is running on a satellite in orbit

Tue, 13 Oct 2026 06:00:00 GMT

Orbitworks and Loft Orbital say the first satellite of a planned 50-satellite AI constellation is in orbit after an Oct 1 SpaceX rideshare. It flies two NVIDIA Jetson AGX Orin modules and runs Mistral Ministral 3 (3B and 8B) onboard to summarize acquisitions and answer operator queries — alongside nearly 20 other models. Alert in seconds instead of dumping raw imagery hours later; first bird of a $1B program. Source: Loft Orbital https://loftorbital.com/orbitworks-launches-first-spacecraft-of-the-largest-ai-powered-satellite-constellation-with-nvidia-edge-ai-onboard/

ScaleDown CDMs: claims 200x faster tools in LangChain

Tue, 13 Oct 2026 06:00:00 GMT

LangChain wired ScaleDown's Summarize, Compress, Extract, and Classify CDMs into Deep Agents. ScaleDown reports CDMs are 200× faster and 467× cheaper than frontier models on these narrow tasks; up to 1M-token inputs; Summarize/Extract/Classify multimodal. On ScaleDown's benches, Summarize was 41× cheaper and Compress 138× cheaper vs an average of GPT-5.5 / 5.4 Mini / Nano. Source: LangChain https://www.langchain.com/blog/building-long-running-agents-with-scaledown-and-deep-agents