Skip to content
Only First-party#Open-source ecosystem
  1. AWS Machine Learning Blog66

    AWS updates AgentCore runtime with 15x faster session starts and 4x more compute

    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.

    Why it matters: AgentCore runtime 的新版本把会话启动、算力上限和内存计费一起调整，为需要长时间运行的智能体提供了隔离与成本之间的新取舍。

  1. vLLM Blog64

    vLLM-Omni benchmarks MiniMax H3-Image on Vera Rubin: 0.87s 4K generation and 6,115 images/hour

    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.

  2. vLLM Blog60

    vLLM adds day-0 support for Thinking Machines' Inkling 2 Small

    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.

    Why it matters: vLLM 在发布当天即支持 Inkling 2 Small，并给出内核融合、序列并行和 Gluon 解码内核的具体优化数据，可供部署大 MoE 模型的团队参考。

  1. NVIDIA Blog60

    NVIDIA Contributes OpenShell Agent Runtime to CNCF and Adds DSX Kubernetes Tools

    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.

  1. Goodfire Research62

    Goodfire studies Qwen3.6-35B-A3B neurons as an interpretable basis

    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.

  1. OpenClaw Blog69

    OpenClaw releases Multiplayer for shared agents and context

    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.

    Why it matters: OpenClaw 把多人共享同一 Agent 与上下文做成免费开源功能，为团队协作中反复重建上下文的问题提供了一种可迁移的做法。

  1. vLLM Blog73

    Buoyant releases Kinference, an open source inference control plane for vLLM and Kubernetes

    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.

  1. Anthropic Newsroom71

    Anthropic donates Agent Skills standard to the Agentic AI Foundation

    Anthropic is donating Agent Skills, the open standard for giving AI agents specialized knowledge and workflows, to the Agentic AI Foundation (AAIF) under the Linux Foundation. More than 70 agent platforms including ChatGPT, Cursor, GitHub Copilot and VS Code have integrated the standard since it was published in December 2025, with 1 million skills published and nearly 280 million installed. The donation covers the Agent Skills name and logo, the GitHub organization and the agentskills.io domain, with engineers from Anthropic, Google, OpenAI and other companies maintaining it as individuals under Lead Maintainers David Soria Parra and Mahesh Murag.

    Why it matters: Anthropic hands its Agent Skills standard to the Linux Foundation's AAIF, the same neutral home as MCP, giving the community a formal say over the spec's roadmap.

  1. Meta AI Research Blog88

    Meta releases Muse Spark Open, an 816B open-weight multimodal model that runs on two Mac Studios

    Meta released the weights for Muse Spark Open, an open-weight natively multimodal model with 816B total parameters (42B active) and a 1M-token context window, positioned for coding and agentic work. It is the first model in its class to run on a desktop using two Apple M5 Ultra Mac Studios with the MLX framework, and it is available through API partners including Baseten, Fireworks AI, Groq, Modal, Nebius, OpenRouter and Together AI, with weights on Hugging Face, Amazon SageMaker AI, Google Cloud Model Garden and Microsoft Foundry Models. Meta says the model ships with auto-tunable Helion GPU kernels in SGLang, delivering 1.03-3.95x kernel-level speedups on NVIDIA and AMD GPUs and a 3% end-to-end speedup on AMD MI350.

    Why it matters: Meta releases open weights for an 816B-parameter multimodal model that runs on two Mac Studios, giving developers a local option for long-horizon coding and agentic tasks.

  1. Goodfire Research71

    Goodfire trains and deploys probe-based cyber misuse monitors on Kimi K3 and GLM 5.3

    3 and deployed them on a production inference stack, where a cheap probe filters turns and only suspicious exchanges reach an LLM judge. 5% benign-session interruption rate, costs $65 per million turns without judge reasoning and $185 with it, and adds no measurable throughput loss inside SGLang.

    Why it matters: Goodfire shows a probe-plus-judge cascade can monitor agentic cyber misuse at roughly 50x lower cost with no measurable throughput loss, a concrete recipe for serving open models safely.

  1. OpenClaw Blog74

    Microsoft Autopilot is built on OpenClaw, with contributions flowing both ways

    Microsoft introduced Autopilot, a persistent personal agent built on OpenClaw, as part of its new Copilot experience, with a private preview expanding to first customers at the end of this month. Autopilot is the new name for Microsoft Scout, announced earlier this year, and Omar Shahine's team says it is working with OpenClaw and the OpenClaw Foundation on an enterprise-grade runtime. OpenClaw highlights upstream contributions from Microsoft engineers, including policy conformance checks, a native Windows companion, and reliability fixes for background jobs, restarts and duplicate replies.

    Why it matters: Microsoft's Autopilot is built on OpenClaw, and the upstream contributions listed here show how a shared agent runtime improves across platforms.

  1. OpenClaw Blog60

    OpenClaw adds opt-in decision models and a shared Plugin SDK interface

    OpenClaw introduced support for a separately configured decision model that its core and plugins can call through api.runtime.decisions.evaluate in the Plugin SDK, returning a choice, score or probability instead of a chat reply. The feature is opt-in and does not replace the chat model, and the provider packages are still awaiting a supporting release. The TypeSafe adapter supports hosted Jev and a local System One server such as Jared Palmer's Kev, and the evaluation tool is moving into OpenClaw core as decision_evaluate.

    Why it matters: OpenClaw adds an opt-in decision model layer that returns typed answers from application code, with a shared Plugin SDK interface so plugins can use it without per-provider integrations.

  1. MiniMax Blog71

    MiniMax releases Music 3.0, an open-weights music generation model for full songs up to five minutes

    MiniMax introduced Music 3.0, an open-weights music generation model that composes, arranges and produces a complete song of up to five minutes from a creative concept and optional lyrics in a single generation. The pipeline pairs an 8B Global LLM initialized from Qwen3.5-8B with a 0.6B Local LLM, fuses their continuous hidden states to condition a 2.4B flow-matching module and a 123M Flow-VAE, and uses eight-layer RVQ plus Structured Captions that describe emotion, instrumentation and vocal delivery over time. MiniMax says this yields more accurate interpretation of creative intent, fuller arrangements and clearer, more natural vocals.

    Why it matters: MiniMax's architecture breakdown shows how an 8B global LLM paired with a 0.6B local model and flow-matching stack targets the long-song coherence problem.

  1. Modal Blog75

    Moonshot's Kimi K3 now available on Modal with day-zero vLLM support

    Modal made Moonshot's Kimi K3 available on its platform on release day, offering the 2.8 trillion parameter multimodal model with a 1M token context window through a token-priced Shared API and dedicated Auto Endpoints. Modal says a custom DFlash speculator tuned to K3's architecture raises interactivity from 100 to 460 tokens per second and throughput from 800k to 1.5 million TPM per GPU on agentic workloads. The model uses a mixture-of-experts design with 16 of 896 experts active per token, and Moonshot contributed a new prefix-caching implementation to vLLM ahead of the release.

    Why it matters: Modal details how it brought Moonshot's 2.8T-parameter Kimi K3 to its platform on release day, including a custom DFlash speculator that lifts interactivity from 100 to 460 tokens per second.

  1. Modal Blog60

    Modal launches Auto Endpoints for self-serve, OpenAI API-compatible LLM inference

    Modal launched Auto Endpoints, a self-serve service that deploys production-grade LLM inference as an OpenAI API-compatible endpoint backed by a Modal App the user can see and control. Each supported model ships with a starting deployment tuned from Modal's own benchmarking, using engines like SGLang and the DFlash speculative decoding drafter from Z Lab, and users can inspect GPU choices, engine flags and inference metrics such as TTFT, ITL and speculative decoding acceptance length. Modal also released Modal Servers from beta for regionalized routing with 5ms overhead, and said it generates endpoint deployments with an internal agentic system still monitored by human engineers.

    Why it matters: Modal's Auto Endpoints expose the serving stack, engine metrics and benchmark recipes that managed inference providers keep hidden, giving teams a self-serve path to owning open-model inference.

  1. Modal Blog60

    Modal releases DFlash speculators for Qwen 3.5 and 3.6, claiming 5-20% extra speedup

    Modal released a set of DFlash draft models for the Qwen series on Hugging Face, including Qwen 3.6 35B-A3B-DFlash and Qwen 3.5 variants from 4B to 122B-A10B, plus a DFlash speculator for Qwen 3.5 397B-A17B built with SGLang. The company says the new drafters add 5-20% speedup over existing DFlash baselines and can run Qwen 3.5 122B-A10B at over 1000 tok/s at concurrency 1 on a B200 node, versus 250 tok/s without speculation. The post argues speculative decoding delivers 2x-3x gains where kernel optimization yields small percentage points, and presents a roofline model for estimating speedup from acceptance length.

    Why it matters: Modal explains why it bets on speculative decoding and releases DFlash draft models for the Qwen series, with measured speedups and a roofline model for estimating gains.

  1. MiniMax Blog87

    MiniMax M3 released with 1M context, frontier coding and native multimodality

    MiniMax released M3, an open-weight model it says is the first to combine frontier coding and agentic performance, a 1M-token context window, and native image and video input. M3 uses MiniMax's new MSA sparse attention architecture, which the company reports cuts per-token compute at 1M context to 1/20 of the previous generation and speeds prefill by more than 9x and decoding by more than 15x. It scores 59.0% on SWE-Bench Pro, 66.0% on Terminal-Bench 2.1 and 74.2% on MCP Atlas, and is available through MiniMax Code, the Token Plan and the API, with weights and a technical report promised within 10 days.

    Why it matters: MiniMax M3 combines a 1M-token context, frontier coding and agentic scores, and native image and video input in one open-weight model, with MSA cutting per-token compute at long context.

  1. Modal Blog78

    Modal 上线 GLM-5 免费端点并发布自部署与前端接入指南

    Modal 推出 GLM-5 的免费 OpenAI API 兼容端点，可用至 4 月底，限单个并发请求，同时发布在单节点 8 张 B200 上用 SGLang 部署的可复现代码。

    Why it matters: Modal 为 Z.ai 的 GLM-5 提供免费 OpenAI 兼容端点，并给出在 OpenCode、OpenClaw、Claude Code 和 Vercel AI SDK 中接入的具体配置。

You are all caught up