Skip to content
#Deployment / engineering
  1. Hacker News · AI76

    AgentRun Open-Source Release: Adding Self-Improvement and Durable Execution to Existing Agent Harnesses

    Parcha Labs has open-sourced AgentRun, an agent platform that layers self-improvement, durable execution, and a console on top of Claude Code, Codex, OpenCode, and pi, under the Apache-2.0 license.

    Why it matters: AgentRun layers self-improvement, durable execution, and a console on top of existing harnesses such as Claude Code and Codex, and provides a concrete case where code review costs dropped from $6,451 per month to 18 cents per run.

  2. 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。

  3. SemiAnalysis85

    Nvidia and the World Without Jensen: What a Post-Founder Nvidia Looks Like

    SemiAnalysis argues that Nvidia was built around Jensen Huang, who has left the CEO role after 33 years, and that his departure weakens the company's long-term bets, chip allocation deals and personal supply chain lockups. The piece says Nvidia had about 60 direct reports under Huang and lacks a deep bench of successors, with Colette Kress as interim CEO and Ian Buck and Jonah Alben floated as internal technical candidates. It also ties the timing to Washington: more than 1.5GW of restricted Nvidia chips are estimated to have reached China, and the Remote Access Security Act, which passed the House 369-22, is up for a Senate vote.

    Why it matters: SemiAnalysis examines how Nvidia's founder-centric structure, supply chain deals and China policy could change after Jensen Huang's exit, and why export control legislation is moving now.

  4. AWS Machine Learning Blog60

    AWS open-sources UAEF, a framework-agnostic evaluation system for AI agents

    AWS has open-sourced UAEF (Universal Agent Evaluation Framework), a framework-agnostic system that uses adapters to normalize agent traces from LangGraph, LangChain, Strands Agents, Amazon Bedrock AgentCore and Langfuse into one canonical AgentTrace, then scores them with 36 built-in metrics across 7 dimensions. Results are stored per experiment in Amazon DynamoDB and Amazon S3, so teams can compare agent versions against a baseline, flag regressions with a default 5 percent threshold, and generate statistical reports plus LLM-written insights. The post says customers across 12 organizations in telecom, healthcare and financial services used UAEF before release, and the Apache 2.0 code, Python library, CDK deployment and per-framework notebooks are in the awslabs/uaef GitHub repository.

    Why it matters: The AWS team has open-sourced UAEF, which uses adapters to unify agent trajectories from different frameworks into a single format, then scores them across 7 dimensions using 36 metrics, providing a reusable evaluation layer for comparing agent versions across frameworks.

  1. The Decoder80

    Anthropic says AI cut its Claude port to AMD GPUs from months to three days

    Anthropic says it ported Claude to AMD GPUs in three days with almost no human intervention, using its frontier model Mythos to write nearly 100 percent of the kernels itself. Co-founder Tom Brown says Anthropic now trains and runs Claude across AWS Trainium, Google TPUs, Nvidia GPUs and AMD GPUs, treating them as one interchangeable compute pool. The figures come from an investor presentation and have not been independently verified; Anthropic cites three days versus the six to twelve months traditional engineering needs, measured on prompt processing with AMD MI355X GPUs.

    Why it matters: Anthropic's claim that its own model ported Claude to AMD GPUs in three days offers a concrete test of whether AI can loosen CUDA's grip on chip choice.

  2. 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.

  3. 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。

  4. vLLM Blog72

    vLLM serves DeepSeek-V4.1-Flash on Vera Rubin NVL72 at 58x GB300 NVL72 throughput

    vLLM reports serving DeepSeek-V4.1-Flash on a single NVIDIA Vera Rubin NVL72 rack at 1.7M tokens/s per GPU with 387 tokens/s per user, 58x the GB300 NVL72, peaking at 2.52M tokens/s and 1,810 tokens/s for a single user. The gains come from kernels rebuilt for Rubin's split memory domains and FP4 tensor cores, a specialized TP8 decode path, and rack-wide KV and Engram pools shared over NVLink. With prefixes pooled across the rack and Dynamo workload-aware routing, the prefix hit rate holds at 96.5-97% even at high concurrency.

    Why it matters: vLLM's third DeepSeek-V4.1-Flash post details how Rubin-specific kernels and rack-wide memory pooling lift throughput 58x over GB300 NVL72, offering a concrete reference for large-scale serving optimization.

  5. Google Blog · AI78

    Google launches Gemini Decisions API and Gemini Decision 1 model for fast closed-set decisions

    Google introduced the Gemini Decisions API, powered by the new Gemini Decision 1 model (gemini-decision-1) and available through the Gemini API in Google AI Studio. A single multimodal input of up to 256K tokens can be checked against up to 1,000 yes/no, multiple-choice and rubric-scoring questions in one forward pass, returning probabilities and confidence scores at 150-250 ms end-to-end latency for $0.10 per 1M input tokens with no output-token charges. Google says the model leads its benchmark suite with the lowest calibration error and is nearly 5x cheaper than Gemini 3.5 Flash-Lite with the Interactions API; internal uses include model routing in Gemini Notebook, wireframe generation in Stitch and adversarial-content checks for MCP tools in Search.

    Why it matters: Google's new endpoint scores up to 1,000 multimodal yes/no, choice and rubric questions in one pass at 150-250 ms, giving agent routing and policy checks a cheaper alternative to generative calls.

  1. Simon Willison78

    Anthropic IPO roadshow video details multi-chip Claude training and serving strategy

    Anthropic kicked off its IPO roadshow yesterday, and its roadshow video disclosed previously unreleased details about how the business works, including a section on compute strategy with co-founder Tom Brown. Brown said Anthropic trains and serves Claude across AWS Trainium, Google TPUs, NVIDIA GPUs and AMD GPUs, and that when the company began working with AMD GPUs this year, a system called Mythos got Claude running on AMD in three days with basically no human intervention, porting or writing almost 100% of the kernels. He also described compute multipliers as a second driver of compute advantage, saying each multiplier stacks on the ones before it and that Claude may eventually discover them entirely by itself.

    Why it matters: Anthropic's IPO roadshow video details its multi-chip training and serving strategy across four chip families, including a Claude-assisted AMD port completed in three days.

  2. OpenRouter Announcements65

    OpenRouter measures the context cost of MCP, CLI and skills on one GitHub task

    OpenRouter measured the same GitHub task across the GitHub MCP server, the gh CLI and a gh skill on Claude Sonnet 5.5, GPT-6 Luna and Gemini 3.8 Flash. On Sonnet 5.5 all three answered correctly, but the CLI used 1,421 input tokens, the skill 3,315 and the MCP server 59,417 with its 46 default tools, whose definitions alone add 19,436 tokens per request. Scoping the server to two tools cut the task to 23,228 tokens, capping results at 4,000 characters brought it to 5,357, and Tool Search cut the up-front cost of 46 tools to under 600 tokens, though it halved the whole-task cost on Sonnet 5.5 while raising it on Claude Haiku 5.5.

    Why it matters: OpenRouter's measured comparison shows how much context an MCP server, a CLI and a skill each consume on the same GitHub task, and where trimming tool results or deferring definitions changes the bill.

  3. OpenRouter Announcements60

    OpenRouter tests what fp8 and fp4 quantization do to LLM quality and price

    OpenRouter ran 280 exact-answer tasks against six endpoints of Qwen3.8 27B on October 20, 2026 and did not detect an accuracy difference between fp16, fp8 and fp4, with all six scoring between 98.3% and 99.2%. Across 1,455 endpoints, 59.5% report no precision at all and fp8 is the most common labeled format at 23.6%. Lower precision was cheaper for most models but not all, and provider choice moved price more than precision did, with the priciest same-precision endpoint a median 1.72 times the cheapest.

    Why it matters: OpenRouter measured fp8, fp4 and fp16 endpoints of one model across 240 exact-answer tasks and found no detectable accuracy gap, while price differences tracked the provider more than the precision.

  4. LangChain Blog60

    LangChain explains how Deep Agents manages context without keeping everything in the window

    LangChain details how its open source Deep Agents harness uses middleware to fight context rot on long agent runs. Tool results over roughly 20,000 tokens and user messages over 50,000 tokens are offloaded to files with only a preview shown, while summarization condenses older messages once usage hits 85% of the context window. Skills, subagents and a markdown long-term memory file keep the main agent's window small and let memory carry over across models and providers.

    Why it matters: LangChain explains how its open source Deep Agents harness uses middleware to offload, summarize and isolate context, offering a reusable pattern for agents on long tasks.

  5. 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. Hugging Face Blog60

    Hugging Face explains decision models and their support in Transformers and llama.cpp

    Hugging Face published an explainer on decision models, which take a state plus typed questions and return probabilities over predefined answers instead of generating text. The post details the choice, score and noul primitives, the confidence formula, and new support through Transformers' DecisionPipeline and AutoModelForDecision, a shared /v1/systemone endpoint in Transformers and llama.cpp, a Hub tag for discovery, and fine-tuning with the Trainer API. It notes that the same endpoint also works with compatible autoregressive LLMs by scoring answer logits in a batched forward pass.

    Why it matters: Hugging Face explains how decision models turn classification into a typed, non-generative interface and shows how to serve and fine-tune them with Transformers and llama.cpp.

  2. OpenRouter Announcements72

    OpenRouter benchmarks 15 vision models and Mistral OCR on scanned documents

    OpenRouter tested 15 vision models, Mistral OCR and Tesseract on eight self-made degraded scans, and Claude Opus 5.5 was most accurate with 262 of 264 key values correct at $16.42 per 1,000 pages, while Mistral OCR was best value at 258 of 264 for $2.02. On a faded receipt, GPT-6 Luna and GPT-6.1 Sol invented card digits, a server name and wifi passwords, errors that character error rates do not reveal. The post includes code to transcribe scans, turn reasoning off, cross-check two models from different labs and run Mistral OCR through the file-parser plugin.

    Why it matters: A hands-on benchmark of 15 vision models plus Mistral OCR and Tesseract on eight degraded scans, with runnable code for cross-checking two models to catch fabricated values.

  1. Anthropic · Claude.dev60

    Anthropic guide on choosing among Claude Fable 5.2, Opus 5.5, Sonnet 5.5 and Haiku 5.5

    Anthropic published a guide recommending developers start most coding and knowledge work on Claude Opus 5.5 at its default medium effort level, since it costs half as much per token as Fable 5.2. The guide maps Sonnet 5.5 to well-scoped everyday work, Haiku 5.5 to high-volume classification, routing and subagent tasks, and Fable 5.2 to the hardest coding, finance and science projects or long unattended runs. It also advises adjusting effort level before switching models, and notes that on SWE-bench Pro, Opus 5.5 at low with failures re-run at high passed about 97% of tasks for about $0.17 each, against 95.3% for $0.29 with every task at high.

    Why it matters: Anthropic's own guide maps each Claude model and effort level to concrete task types, with per-task cost and benchmark comparisons for choosing between them.

  2. Zyphra Research78

    Zyphra releases ZONOS3-Mini, an Apache 2.0 zero-shot voice-cloning TTS model that runs on-device

    Zyphra released ZONOS3-Mini, a sparse MoE text-to-speech model with 100M active and 800M total parameters under Apache 2.0, trained on 112 NVIDIA H100s in 2.5 days. It reaches under 570 ms time-to-first-audio in the browser on a Pixel 9 and under 150 ms on an M3 MacBook Air, with runtimes for WebGPU, SGLang, vLLM and GGML. The model adds control over identity, accent, quality, mood and delivery, including word-level steering, and Zyphra reports it ranks first among open models on its new ZTTS2-Eval benchmark.

    Why it matters: Zyphra's compact MoE TTS model pairs zero-shot voice cloning with word-level delivery control while running in real time on phones and laptops.

  3. Google Developers Blog60

    Google Cloud MaxText adds context parallelism for long-context training on TPUs

    Google Cloud released long-context training support in MaxText v0.2.4, its open-source JAX training framework, adding four context-parallel strategies on Cloud TPU: All-gather, A2A (DeepSpeed Ulysses), P2P (Ring Attention) and the A2A + P2P hybrid (USP). Benchmarks with Llama 3.1 8B reached 70% MFU on TPU 5p at sequence lengths up to 4 million tokens, and context extended to 14 million tokens on both TPU 5p and TPU 7, with TPU 7 holding 50% MFU below 6 million tokens. The release is available on PyPI, in Docker images and on GitHub main, with sample recipes for TPU 7 and TPU 5p.

    Why it matters: MaxText's four context-parallel strategies on Cloud TPU, with benchmark MFU figures for Llama 3.1 8B, give teams a concrete reference for choosing how to train at multi-million-token sequence lengths.

  4. Anthropic Research71

    Anthropic tests Claude Fable 5.2 on real-world electronics design

    Anthropic reports that Claude Fable 5.2 designed three electronic devices, a speakerphone, a voice-assistant hub and a handheld virtual pet, from minimal prompts, and that Diode Computers built working prototypes from its design files without circuit fixes. On Diode's internal PCB Gym benchmark, Fable 5.2 scored 69% versus 25% for Fable 5, and it reached 83.5% on the public BenchCAD 2.1, nearly double Claude Fable 5.1. The devices remain prototypes that have not been through electrical validation, antenna tuning, thermal testing or certification, and Claude was not connected to test equipment or manufacturing machines.

    Why it matters: Anthropic's own account of Claude Fable 5.2 designing three working electronic devices shows how far models have moved from partial PCB layouts to end-to-end hardware prototyping.