llama.cpp has released v1.0.0, and ggml has also released v1.0.0. Going forward, both projects will indicate incompatible changes to the public C API and ABI through major version number changes. The project team states that as local AI and privacy needs grow and downstream applications increase, stable library interfaces have become critical, and the boundaries for breaking changes will be stricter in the future.
OpenClaw is bringing its three-day Claw Hack hackathon and ClawCon conference to Shenzhen on December 4–6, 2026, following its Shanghai event. Claw Hack spans software and hardware tracks, with M5Stack, Seeed Studio's reBot Arm B601 and Pamir hardware, and applications close November 25. ClawCon runs December 6 with talks on open-source AI, open-weight models and running agents locally and in the cloud.
The TimeNet Challenge is an online hackathon that asks teams of 2–4 to onboard a time-series dataset in TimeNet format and train or fine-tune a Temporal AI Model such as OpenTSLM for tasks like forecasting, classification, anomaly detection or reasoning. Round 1 launched November 11 and closes November 22 at 11:59 p.m. PST, with more rounds to come. Nebius offers $100 or $500 vouchers per team per round, and winning teams receive prize credits starting at $25,000 from AWS and Google Cloud.
ArtCraft founder and CEO Brandon Thomas presented at the AI Engineer Code Summit in San Francisco a manifesto for AI-based open source, saying the company has released 17 open source versions of popular software tools, including a Photoshop alternative called PhotoCraft built in Rust and a PDF tool called PdfCraft.
The Open Source Initiative has launched the Open Source AI Salons, a series of community events examining how the Open Source AI Definition (OSAID 1.0) applies in practice, with the first gatherings held in Raleigh and Barcelona. Participants agreed the four freedoms must be preserved and that training data transparency remains the biggest bottleneck, as Hugging Face's model count grew from roughly one million in 2024 to over three million today.
Bright Machines now gives customers live, governed dashboards of factory data from air-gapped plants without opening any inbound access. It captures PostgreSQL changes at the edge with Debezium, writes them as Apache Iceberg, pushes them out over one auditable outbound path, and reads them in place in Databricks via Unity Catalog federation. The first customer-visible dashboard went live about two months after kickoff, with committed rows queryable in one to two minutes and dashboards refreshing every ten minutes.
Amazon announced the public availability of SGLang-Neuron, an out-of-tree plugin that runs SGLang on AWS Trainium through an AWS Deep Learning Container, released by RadixArk with Annapurna Labs. Users keep the same SGLang server interface and generation APIs, while the plugin supplies Trainium-specific models, attention, MoE and a graph runner built on native PyTorch and NKI kernels compiled with torch.compile. It also supports PD disaggregation, transferring the prefill KV cache to a decode server via NIXL over EFA, and the walkthrough serves openai/gpt-oss-120b on a Trainium3 EC2 instance with the pytorch-inference-sglang-neuronx container.
CopilotKit released OpenBot 1.0 on November 9, an MIT-licensed, self-hostable agent workspace where teams bring their own models, agents and infrastructure. Every browser click, file write, MCP tool call and component passes through a governance gateway that evaluates a CEL policy and writes the audit row before acting, with deny rules running first and broken rules failing closed. The release adds Mac and Windows desktop apps, a Learning feature that turns a recorded task into a reusable skill, multi-account support through Composio and Arcade, and Bot-to-Bot orchestration; personal use is free and OpenBot for Business costs up to $29 per active user per month.
Nvidia CEO Jensen Huang resigned Thursday after admitting to an affair with an alleged Chinese spy, leaving open whether Nvidia will keep backing open-weight AI. Huang had pushed the Nemotron open-weight model family, backed startups like Reflection AI and Thinking Machines Lab, and organized an open letter signed by 25 companies urging Washington not to over-restrict open models. Nvidia has named CFO Colette Kress as interim CEO and has yet to finalize its $13 billion deal to acquire Hugging Face.
AWS announced updates to Amazon Bedrock AgentCore runtime, its managed compute layer for agents, with sessions now starting up to 15x faster, scaling to 4x more compute (up to 8 vCPUs and 32 GB, with container images up to 10 GB) and running up to 68% more efficiently, while each session stays in its own hardware-isolated Firecracker MicroVM. Sessions start in under 2 seconds regardless of image size or concurrency, idle agents scale to zero, and agents can mount and save work to a filesystem such as an Amazon S3 bucket with FUSE support so data stays inside the trust boundary. The runtime is available today in AWS Regions across North America, South America, Europe, and Asia Pacific.
Hugging Face CEO Clément Delangue said he will keep fighting for open-source AI "harder than ever" after Nvidia CEO Jensen Huang's departure, calling it a "really sad day." Nvidia announced in September a $12.9 billion plan to acquire Hugging Face, a deal the U.S. Department of Justice has been inquiring about over competition concerns and which is still scheduled to close in the first half of 2027. Huang had been a prominent advocate for open-weight models, including a July policy letter signed by 25 companies urging the U.S. government not to over-restrict them.
AWS has published a catalog of 23 production-ready, open source generative AI assets from its Generative AI Innovation Center, now part of AWS Frontier AI Engineering & Services. The building blocks cover RAG pipelines, document processors, agent scaffolding and evaluation frameworks, and AWS says they can cut development from months to weeks. DoorDash reported a 50x increase in testing capacity and Ricoh a 7-fold throughput increase after building on the assets.
vLLM supports Thinking Machines' Inkling 2 Small on day 0, a multimodal text-and-vision MoE with 594B total and 17B active parameters and up to 1M tokens of context. The integration adds kernel fusion with PDL, sequence parallelism for prefill that cuts an 8K-token prefill step by 22% at TP4 and 26% at TP8, and a Gluon relative attention decode kernel that drops a batch-1 decode step at 100K context from 15.1 ms to 4.8 ms. vLLM reports up to 630 TPS/user at TP4 on GB200 with acceptance length 3.5, and also supports LoRA and RL via SkyRL.
vLLM-Omni brought MiniMax H3-Image to NVIDIA Vera Rubin and measured 0.87-second median latency for native 4096×4096 text-to-image on four GPUs, plus 6,115 images/hour in a separate throughput configuration. H3-Image is an open-weight text-to-image and editing model that pairs a Qwen3-VL encoder with a 33.1B dense DiT and an image VAE, handling both tasks in six sampling steps with up to 16 reference images. The team reports MXFP8 DiT GEMMs, Ulysses SP4, AdaLN caching and a VAE decoding change that cut decode time from 2.635s to 0.213s, and lists recipes for AMD, Intel, Huawei Ascend and Moore Threads hardware.
Why it matters: vLLM-Omni's benchmark shows MiniMax H3-Image running on Vera Rubin at subsecond 4K latency, with per-hardware configurations and a cost model for serving economics.
Epoch AI researchers report that in offline sandbox experiments, the open-weight model GLM 5.3 autonomously compromised six recent versions of Firefox, a browser with over 200 million active users, for about $1,000 per exploit by reverse-engineering attacks from patches that had not yet reached users. The model built webpages that could read files from a visitor's computer and send them to an attacker without human input; the team could not replicate this with Chrome. The same brief notes that Anthropic's Claude Fable 5.2, released November 9th, set a new Epoch Capabilities Index record of 174, above GPT-6 Astra's 169.
Mastra added private model gateways that let agents route requests to self-hosted open-weight models, referenced by a prefix like private/ollama/gpt-oss:20b. Self-hosting keeps sensitive data inside the network and replaces per-token costs with infrastructure costs, and setting AUTO_BLOCK_EXTERNAL_PROVIDERS=true hides external providers in Studio. The feature requires @mastra/core@1.82.0 or later.
Modal is one of three companies, alongside Ant Group and Tines, joining gVisor's maintainers as Google donates the project and its trademarks to CNCF under a maintainer-led governance model. Modal says it launches upwards of 40 million gVisor containers a day and has contributed upstream since 2023, including nvproxy GPU support, an RDMA device proxy called rdmaproxy, and tmpfs and memory-allocator optimizations. The company says it is working on multi-GPU memory snapshots and GPUDirect Storage support for gVisor.
vLLM has expanded Encoder/Prefill/Decode (EPD) disaggregation for multimodal serving, adding metadata-only requests, a dedicated encoder execution path, remote embedding transfer via NIXL and Mooncake, cross-encoder output reuse, and orchestration through vLLM Router, SMG, and llm-d. Core capabilities including MiMo and XPU support ship in vLLM v0.32.0, while metadata-only support for DeepSeek V4.1 Flash is available on main via #62802. In a DeepSeek V4.1 Flash test on GB200, EPD delivered 6.6–9.5% higher per-GPU throughput and 44–47% lower P90 TTFT than an eight-GPU PD baseline at concurrency 64–128, but had lower per-GPU throughput at concurrency 16.
Meta detailed TLX (Triton Low-Level Extensions), an extension to Triton that adds explicit warp-group and cluster-level orchestration while keeping Triton's tensor programming model. TLX is already running in production across Meta, with 20-25% end-to-end MFU on thousands of NVIDIA Blackwell GPUs for GEM training, up to two-thirds lower compute-intensive net latency for RecSys inference, and a distributed grouped GEMM kernel deployed in large-scale LLM pretraining. Meta says TLX targets NVIDIA Hopper and Blackwell plus AMD Mi300 and Mi350x, and is working on more architectures, automatic warp-group inference and a torch.compile backend.
NVIDIA is contributing OpenShell, its open source runtime for securing and governing autonomous agents, to the Cloud Native Computing Foundation as a Sandbox project, and is also contributing to Google-led Agent Substrate. NVIDIA announced three new DSX AI Factory Platform components: NVIDIA Kubernetes Engine and NVIDIA Kubernetes Recipes, now in public preview, and NVIDIA Snapshot, which restores checkpointed replicas in seconds without model reload or warmup. Crusoe tested GLM-5.3 on Crusoe Cloud and cut workload ready time from roughly eight minutes to 31 seconds, while Saturn Cloud reduced vLLM startup from between two and 8.5 minutes to six and 34 seconds.
Why it matters: NVIDIA contributes its agent security runtime OpenShell to CNCF and adds Kubernetes tooling for GPU AI factories, with early Snapshot results cutting inference startup from minutes to seconds.
The Pragmatic Engineer interviews Sean Allen, a distinguished engineer at Antithesis and core contributor to the Pony language, on software correctness and supply chain security. Allen argues that strong type systems in the ML language family catch more errors at compile time than ALGOL-family languages, and that manual testing is the worst way to ensure correctness. He also says supply chain attacks have no single simple fix, since robustly defended languages like 'E' are unpopular while popular languages assume a trustworthy ecosystem.
Stripe built Kai, its company-wide Knowledge AI Platform, on LangChain's open-source Deep Agents harness, with one engineer shipping the initial version in a single week. The stack layers a Stripe-specific harness and configuration layer over Deep Agents, using filesystem middleware backed by S3, sandbox middleware for code execution, summarization middleware for long sessions, and skill-based dynamic tool loading across 500+ internal MCP tools and 1,000+ skills. Kai grew from 296 to over 5,000 users in roughly four weeks and is now used weekly by 83% of Stripe across more than 60k sessions.
Meta has built a vendor-neutral accelerator plugin architecture in lldb-server that lets LLDB debug applications spanning CPUs and GPUs in a single session. The design represents the CPU and GPU as separate LLDB targets, each with its own architecture, threads, registers and address space, while a plugin translates vendor device debugging APIs into LLDB concepts. GPU support activates only when the app uses a supported GPU runtime, so CPU-only debugging keeps the existing native path without added overhead.
Goodfire Research published a study arguing that individual MLP neurons in Qwen3.6-35B-A3B are more interpretable than assumed, based on sparsification, attribution tools and research agents. Ablating the 92.2% weakest MoE neuron activations per forward pass raised language model loss by only 2.9%, and 69% of neuron poles had a single mode responsible for at least 90% of their causal effect. Natural-language neuron descriptions let a judge model predict firing with 95% accuracy against random text and recover 73% of ablation damage through simulation.
Why it matters: Goodfire's study of Qwen3.6-35B-A3B neurons argues that strong activations alone carry most of the model's computation, offering a concrete alternative to feature-dictionary interpretability.
Preferred Networks released PLaMo 3.0 Fin, a finance-specialized small language model built on the PLaMo 3 NICT pretraining model from joint research with NICT and further trained on Japanese financial data, publishing open weights on Hugging Face under the PLaMo Community License. The model adds step-by-step Reasoning with reasoning traces output in Japanese, plus tool calling for web search, calculation and internal database access, and can run on a single GPU in closed cloud or on-premise environments. PFN says the instruction-following training for Reasoning and tool use was carried out on its own MN-Core 2 AI chip, with code porting completed within a week and training speed comparable to a GPU environment.
Artificial Analysis launched Endpoint Accuracy Index v1.1, which tightens AA-LCR-25 confidence intervals and adds coverage for GLM-5.3 Flash and Muse Spark Open. Across 103 serverless endpoints benchmarked, 24 score statistically below the self-hosted reference, with causes including output token caps, reduced reasoning, tool call formatting errors and request timeouts. Raising AA-LCR-25 to 25 repeats per question narrows its average 95% confidence interval from ±4.7 to ±2.6 points, and the composite index interval from ±5.5 to ±4.3 points across the 45 endpoints scored under both versions.
Apple began developer testing of Qwen inside Apple Intelligence in China this week, ahead of a planned December rollout with iOS 27.3, bringing Qwen into Siri, Writing Tools and Visual Intelligence on an opt-in basis. Tencent's November 4 Hy4 release adds visual reasoning and computer use, while Baidu, Moonshot, Ant and Meituan also published or launched new models.
vLLM has migrated its CUDA build to the LibTorch Stable ABI, and as of v0.33.0 it publishes LibTorch ABI Stable CUDA wheels, so a new PyTorch wheel can be imported without rebuilding vLLM. The team moved in-tree sources and seven LibTorch-dependent dependencies, including FlashMLA, DeepGEMM and FlashAttention-2/3, onto stable APIs, using torch-abi-audit in CI to confirm no unstable symbols remain. The post notes tradeoffs such as a limited stable API subset and small host-side overhead, and says CPU and ROCm builds are next.
OpenClaw rebuilt how it discovers models so new ones like GPT-6.1 Sol appear without shipping a release, pulling lists from a self-updating catalog at catalog.openclaw.ai (rebuilt every four hours, about 1,000 models), each account's own provider model list, and a handshake with a newer Codex on your PATH. The catalog switches live on a running Gateway, and model lists no longer shrink after provider timeouts or catalog swaps.
Newcomer and Felicis released Emergent, a list of 29 seed-stage or earlier startups nominated by current and former founder speakers from the Cerebral Valley AI Summits as having generational potential. Each company is capped at a seed round and paired with a blurb from its nominator. The largest cluster builds tools to run and control agents, with hardware, open source, and AI-native services also prominent.
SemiAnalysis released the first long-context, multi-turn agentic inference results on InferenceX/AgentX for Google TPUv7 Ironwood across four models, finding it competitive on performance per dollar or state-of-the-art in several cases. On GLM-5.2 in FP8, TPUv7 delivered 1.2x to 2.7x the tokens per dollar of B200, B300 and MI355X from 60 to 300 tok/s/user, while FP4 GPUs still led on GLM-5.3 because TPUv7 lacks native FP4 support until TPUv8 in Q1 2027. The piece also argues NVIDIA's bureaucracy slows driver and ecosystem contributions compared with AMD and Google.
OpenClaw has released Multiplayer, a free and open-source feature that lets multiple people connect to the same OpenClaw agent and workspace, with prompts appearing live across laptops and requests coordinated through a shared queue. In demos at GitHub Universe in San Francisco, two attendees edited the same 3D scene by typing prompts on separate machines without restating context. The OpenClaw Foundation says it has had 50+ people collaborating in the same shared environment and is continuing to expand the feature.
Mastra launched the Custom Harness Program, in which forward-deployed engineers embed with customer teams to co-build tailored agent harnesses. The program includes observability, evals, custom MCP, human-in-the-loop workflows, auth and RBAC, plus early access to the Mastra Enterprise Platform. Mastra cites Workday, which moved off Amazon Bedrock Agents in 31 days, cut 3,000 lines of code and lowered vector-storage costs by 98%.
Ai2 has launched a climate modeling initiative backed by three-year funding from Google.org and technical support from Google Research to develop NeuralGCM+, an open-source model for faster, more reliable climate projections. The project extends Google Research's NeuralGCM, which combines physical equations for large-scale dynamics with AI for smaller-scale processes like cloud formation and rainfall, to simulate atmosphere and ocean together. Ai2 aims to cut regional temperature and rainfall errors over the past 50 years and release models and frameworks with documented workflows for outside researchers.
Buoyant announced Kinference, an Apache v2 open source control plane that routes OpenAI- and Anthropic-compatible inference traffic to vLLM workers on Kubernetes, adding token rate limiting, per-tenant usage and cost tracking, KV-cache-aware load balancing, and YAML inference policies. In the project's public benchmarks, Kinference cut median time-to-first-token by 95% and median end-to-end latency by 45% versus standard Kubernetes service load balancing on one workload, and served 7.4x the goodput on shared-prefix and about 3.5x on multiturn in a capacity test. Policies can route simpler prompts to cheaper models, burst overflow to providers such as Bedrock and Databricks, or keep PII-tagged requests inside the cluster, and per-tenant metrics cover token rates, TTFT, ITL and cached versus uncached prompt tokens.
Why it matters: Buoyant's Kinference adds KV-cache-aware load balancing and YAML inference policies on top of vLLM, with benchmarks claiming large TTFT and goodput gains over standard Kubernetes service routing.
Zyphra introduced Zyphra Hierarchical Attention (ZHA), a sparse attention method that builds a tree of compressed KV-cache summaries and routes queries top-down through it, giving subquadratic complexity with suitable top-p thresholds. Existing pretrained models can be adapted via a short KL-distillation phase using only a few billion tokens; on Qwen3.6-35B-A3B, ZHA adds 81,920 trainable parameters, about 0.00024% of the base model. On RULER-13 at 256K tokens the wide-tree model scored 96.2% versus 96.0% for dense attention, and the deep six-level tree scored 63.4% on LongBench v2 versus 58.0% for dense while attending to about 1.1% of the KV cache; a GPU roofline model projects 32-35x speedup over dense attention at 10M tokens.
Ai2's Olmo 3 received Nextcloud's highest Ethical AI rating, green, which requires open-source training and inference software, a self-hostable trained model, and publicly available training data. Nextcloud bundles Olmo 3 with its llm2 local LLM backend, and the 7B variant was chosen so users without 16 or 32 GB of VRAM still get a good out-of-the-box local experience.
Sakana AI held its "From World Models to Real-World AI" symposium in Tokyo on October 26, 2026, where Chief Scientific Advisor Jürgen Schmidhuber gave a keynote and Sakana AI announced it will develop its own pretrained language model series, Sakana Sawara (tentative name). The first small model, Sawara Small (tentative name), is planned for release in spring 2027 if development proceeds smoothly. More than 400 attendees joined the event, which also featured a conversation between Schmidhuber and CEO David Ha.
Epoch AI reports that GLM 5.3, an open-weight model released by Z.ai in August, autonomously built working exploits from Mozilla security fixes that had been published but not yet shipped to users, compromising all six Firefox versions released between September 17 and October 26. After roughly $100,000 in setup and API costs, each successful attack cost about $1,000 in compute, and the fastest attempt took 12 hours within a 48-hour budget. Similar attacks on Chrome did not succeed, and the researchers say no provider safeguards or refusals were triggered; a separate unpublished vulnerability they found was patched in the October 27 Firefox 159.0 release.
Why it matters: Epoch AI's experiment shows an open-weight model autonomously weaponizing published-but-unshipped Firefox fixes, putting a concrete cost and time figure on the patch gap.
Meta is partnering with five US universities — Caltech, Ohio State's College of Medicine, Louisiana State University, Indiana University and UTHealth Houston — and granting $2 million to support research using its models in medicine, physics, chemistry, biology, computer science and mathematics. Projects include a computer vision system built on Meta's open-weight SAM models to detect tracheostomy tube displacement in infants, multimodal AI to help conversational agents adapt to people with dementia, and open-weight models predicting heart failure risk from routine CT scans and medical records.