Skip to content
Only First-party#Deployment / engineering
November 12Thursday
  1. OpenRouter Announcements58

    OpenRouter explains the self-correcting localization pipeline behind its 7 new languages

    OpenRouter launched beta localization in Simplified Chinese, Spanish, Brazilian Portuguese, German, French, Japanese and Korean, and detailed the self-correcting pipeline that keeps 16,918 English strings translated. The daily loop sends changed strings to GPT-6.1 Sol in batches of up to 40, has code reject broken placeholders and glossary violations, then uses a second call to the same model as a verifier that returns errors by severity for retries. OpenRouter says a full retranslation of the catalog into 7 languages costs about $460, versus about $7,200 at Smartling's published machine translation rate.

  2. OpenRouter Announcements58

    OpenRouter 发布教程：如何阻止 AI Agent 陷入无限循环

    OpenRouter published a guide on stopping AI agents that get stuck retrying a failing tool, using four guards tested on six current models. With doomLoop detection enabled, all six models were blocked on their third identical tool call, and across 18 runs tool calls fell from 314 to 64 while spend dropped from $0.29 to $0.05. The guide adds a loopKey to normalize arguments, stopWhen caps on steps and cost, and a credit limit on the API key that returns a 402 with limit_source openrouter_key_limit.

  3. 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.

  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. Hugging Face Blog36

    Hugging Face Releases Fine-Tuning Guide and Scripts for TRELLIS.2 3D Generation

    Hugging Face published fine-tuning scripts and containers for TRELLIS.2, a 3D generation model, runnable locally or on Hugging Face Jobs. The guide uses LoRA adapters via the PEFT library on only the sparse-structure flow DiT and shape flow DiT, leaving the material stage frozen. On a 188-object footwear subset of the GSO dataset, the pretrained model often rendered incorrect soles, which the LoRA fine-tune targets.

  6. AWS Machine Learning Blog36

    Running OpenAI Codex on Amazon Bedrock: Access Paths and Rollout Guide

    This post shows how to route OpenAI Codex model requests to Amazon Bedrock while reusing existing AWS identity, IAM, Regional, and observability practices. It recommends starting with Native AWS Access for a small least-privilege pilot, with AgentCore Gateway or an LLM Gateway as alternatives, and walks through IAM Identity Center sign-in plus a direct Bedrock Runtime check using GPT-5.6 Sol.

  7. AWS Machine Learning Blog46

    Detecting and recovering from silent data corruption on Amazon SageMaker HyperPod

    A PyTorch DDP training loop on Amazon SageMaker HyperPod gains three checks that catch silent data corruption: a non-finite gradient check, a ring replay audit comparing loss bits and gradient fingerprints across nodes, and a weight fingerprint comparison after the optimizer step. Tested with software-injected faults on 4 ml.p5.48xlarge instances (32 NVIDIA H100 GPUs), the checks name a suspect rank, map it to a node, and let HyperPod replace the node and resume from a checkpoint.

  8. AWS Machine Learning Blog41

    Amazon Bedrock AgentCore Web Search API Released

    Amazon Bedrock AgentCore now offers a standalone Web Search API, letting developers call InvokeWebSearch directly from AWS SDKs or the AWS CLI with a single request and no resources to create. It returns ranked results from an Amazon-operated web index, each with a text extract and, when available, title, URL, and publication time, with domain and date filters. Requests are signed with SigV4 and IAM, using the same permission as the AgentCore Gateway route.

  9. AWS Machine Learning Blog36

    What landed for AI builders on AWS in October 2026: Claude Haiku 5.5, GPT-6.1 Terra, AgentCore CLI 1.0

    AWS's October 2026 roundup adds Claude Haiku 5.5, which Anthropic says costs around 75 percent less than Claude Haiku 4.5 for most tasks, plus OpenAI GPT-6.1 Terra on Amazon Bedrock with up to 1 million input tokens and prices at least 50 percent lower than GPT-6.1 Sol and GPT-5.6 Terra. AgentCore CLI reached version 1.0, and AgentCore Gateway added model fallback and Step Functions workflows as agent tools.

  10. AWS Machine Learning Blog12

    AWS re:Invent 2026 AI/ML Track: Attendee's Guide

    AWS will hold its annual re:Invent conference from November 30 to December 4, with the AI/ML track offering more than 140 unique sessions out of over 2,000 total learning sessions. The guide maps sessions to the AWS AI stack, spanning AI chips (AWS Trainium and NVIDIA GPUs), models and inference on Amazon Bedrock and SageMaker AI, and agent development with Amazon Bedrock AgentCore. Sessions are organized from newcomer content to technical deep dives, and many can be viewed on demand.

  11. AWS Machine Learning Blog28

    Automating AWS Agent Registry updates with a push-based architecture

    AWS outlines an event-driven pipeline that keeps AWS Agent Registry records for MCP servers current without manual updates. AWS CloudTrail captures UpdateAgentRuntime API calls, Amazon EventBridge routes them across accounts, and a Lambda function queries the MCP server's tools via AgentCore Identity OAuth tokens, then diffs and updates the matching registry record. The sample updates existing records only and supports both single- and multi-account setups.

  12. Google Research58

    Google Research uses wearables to build CKM risk profiles and personalized coaching

    Google Research introduced an investigational framework that turns consumer wearable data into a coachable health age and simulates personalized 12-week what-if health trajectories. The SensorWM world model was pretrained on 2.28 million person-weeks of de-identified Fitbit data from roughly 100,000 users and forecasts 20 physiological markers under different step, sleep and cardio routines. In a blinded study, lifestyle coaches chose the counterfactual-grounded plan as best overall in 71.4% of evaluations, and physicians rated 99.2% of plans within safe exercise intensity limits; the authors note the models are research prototypes, not clinical diagnostic tools.

  13. 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.

  14. Amazon Science36

    Using Mathematics to Validate Network Configurations: AWS Tiros for Segmentation Checks

    AWS is using its automated-reasoning tool Tiros to statically validate network segmentation across all modeled paths, rather than relying on packet-based sampling. Tiros converts Cloud WAN configurations from route tables, attachments, and policy documents into logical propositions, then checks that no path exists between segments the policy declares off limits, with analyses run as often as every hour.

  15. Meta Engineering Blog40

    Meta Intelligent Delivery: ML System Allocating Compute for Ads Ranking

    Meta detailed Intelligent Delivery, an ML-driven system that allocates compute across its ads serving stack toward the system-wide optimum, contributing roughly 10% incremental ranking performance on top of model and system advancements from 2017 through the first half of 2026. Its Resource Core Model uses 6B sparse and 20M dense parameters to predict each request's marginal utility, and Meta says it improved identification of high-impact ad requests by about 20% in offline benchmarks. The system also relies on multi-layered allocation mechanisms and increasingly agentic AI to adapt as models and hardware change.

  16. Mastra Blog40

    Mastra Agents Private Model Gateways for Self-Hosted Models

    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.

  17. Claude Blog40

    Anthropic's Claude Tag guide for startups: top Slack channels and use cases

    Anthropic published a Claude Tag (beta) guide for startups, based on conversations with eight high-growth companies about where they deploy the Slack-based agent first. The guide covers five common setups — on-call triage, thread-to-PR, company brain, self-serve data, and reporting — plus a getting-started checklist. Cited results include a PR opened 21 minutes after a Slack mention and 134 pull requests merged from Slack threads at NavigateAI.

  18. Claude Blog56

    Anthropic opens private beta of self-hosted Claude Managed Agents

    Anthropic released self-hosted Claude Managed Agents in private beta, letting developers run the full agent service, including harness, sandboxes, credentials and session history, in their own cloud or on-premise environment. Model requests can go to Anthropic directly, to Amazon Bedrock, Google's Agent Platform or Microsoft Foundry, or through the customer's own gateway, and session history stays in the customer's database. The company says the design keeps the agent loop and vault secrets outside the sandbox, supports allowlists of permitted hosts, and exports a full record of each session over OpenTelemetry.

  19. Figma Blog42

    How 3D transforms and Figma's agent brought a launch video to life

    Figma has released 3D transforms in open beta, letting designers tilt, rotate, and arrange text, images, vectors, and frames in space, with a perspective effect on parent frames for shared spatial compositions. The non-destructive effects stay editable and can be combined with shaders, animated with Figma Motion, or explored via the Figma agent. Figma's Brand Studio used the tools, including an agent-built generative plugin, to produce the feature's launch video.

  20. Claude Blog31

    J.Crew launches Style Advisor AI stylist built with Claude Code

    J.Crew launched Style Advisor, an AI stylist on JCrew.com that recommends complete outfits rather than single products, available now for women's and men's styles in the U.S. J.Crew's team built it with Claude Code using multiple Claude models: Claude Fable 5.1 to plan the stylist reasoning, Claude Opus 5.5 for the prototype, Claude Sonnet 5.5 as the stylist in each conversation, and Claude Haiku 5.5 for quick tasks. Every recommendation is drawn from looks created by J.Crew's own stylists for catalog shoots.

November 11Wednesday
  1. ElevenLabs Blog25

    ElevenLabs Launches Partner Program to Speed Up Enterprise AI Agent Adoption

    ElevenLabs launched its Partner Program with enhanced commercial incentives, new certifications, joint go-to-market opportunities and dedicated technical support. The program covers 550+ active partners in 44 countries, and the partner ecosystem has generated $315 million in pipeline and over $52 million in partner-sourced and partner-influenced revenue this year. ElevenLabs aims for partners to earn $3 in services revenue for every $1 it generates, backed by forward-deployed engineers and a new certified partner directory.

  2. ElevenLabs Blog58

    ElevenLabs Summit NYC: Interaction Engine, ElevenAgents Memory and Routines, ElevenAI app, Partner Program and Marketplace

    At its New York Summit, ElevenLabs announced new models, agents and deployment options for its AI interaction layer, including the Interaction Engine that manages turn-taking and routes each part of a conversation to a cheaper or frontier LLM. ElevenAgents gains Memory, which builds a structured record of returning customers, and Routines, an automation layer for multi-step work before, during and after calls. ElevenLabs also previewed the ElevenAI workplace app with free early access opening today, launched a Partner Program now at more than 550 partners in 44 countries plus a McKinsey alliance, and introduced the ElevenLabs Marketplace; NVIDIA's Nemotron 3.5 Super VL is now available in ElevenAgents alongside 15 MW of new Vera Rubin compute capacity.

  3. ElevenLabs Blog40

    ElevenLabs Launches Marketplace for Enterprise-Ready AI Applications

    ElevenLabs launched the ElevenLabs Marketplace, a curated collection of industry-specific AI and voice applications built on its models by technology partners, with launch partners including Salesforce, IBM, NiCE, Twilio, and Zendesk. Enterprises can explore solutions across financial services, insurance, healthcare, telecommunications, retail, and government, and connect directly with the partners behind them. Starting next year, eligible partner solutions will be purchasable through existing ElevenLabs commitments.

  4. Modal Blog58

    Modal joins gVisor maintainers as Google donates the project to CNCF

    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.

  5. ElevenLabs Blog58

    ElevenLabs and NVIDIA expand partnership to bring voice agents to enterprises and governments

    ElevenLabs and NVIDIA have expanded their collaboration across models, infrastructure and sovereign deployments, bringing NVIDIA Nemotron 3.5 Super VL into the ElevenAgents platform. ElevenLabs plans to add 15MW of NVIDIA Vera Rubin capacity on top of its existing Blackwell Ultra systems, and is powering a voice agent for Poland's Centralna e-Rejestracja appointment system on-premises using NVIDIA Confidential Computing. On the ElevenAgents customer-support benchmark, Nemotron 3.5 Super VL reached leading end-to-end task-completion accuracy at 70% lower cost than other leading open models, according to ElevenLabs.

  6. ElevenLabs Blog22

    How Enterprise Clients Are Putting ElevenLabs AI to Work

    ElevenLabs says enterprises now account for more than 55% of its revenue, with its technology in daily operations at five of the 10 largest tech companies, five of the 10 largest insurers and four of the 10 largest telecoms. At its NYC Summit, clients including Ramp, Customers Bank, Banco Inter, Clover Health and FOX Sports described using ElevenAgents and ElevenCreative for support agents, onboarding, multilingual service and an AI persona of commentator Colin Cowherd in the FOX Sports App.

  7. NVIDIA Blog22

    How AI Helps Recyclers Turn Mixed Waste Into Reusable Material

    NVIDIA highlights four Inception startups using AI and robotics to sort mixed waste at recycling plants. AIRECO's Jetson Orin NX pilot line tops 95% recognition accuracy and targets raising annual capacity from 18,000 to 48,000 tons while cutting operating costs by 85%, and EverestLabs has deployed about 150 vision-guided robots across North America and Australia.

  8. Vapi Blog58

    Vapi launches Agent Loop in early access, a self-improving harness for voice agents

    Vapi launched Agent Loop in early access, with GA planned for early next year, as a self-improving harness that builds a voice agent from a plain prompt, scores every call against a business goal, drafts fixes for the misses, tests them in simulations and ships them to a slice of live traffic before full rollout. The loop runs on Vapi primitives including Simulations and Traffic Splitting, available in the dashboard now, plus Goals, Call Health and Experiments, coming soon. Every Agent Loop agent is a real Vapi assistant, so engineers keep full dashboard and API access, and human approval is required before any change ships.

  9. vLLM Blog58

    vLLM adds encoder disaggregation for multimodal serving with EPD

    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.

  10. 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.

  11. OpenRouter Announcements58

    OpenRouter guide: fixing OpenAI and Claude 429 rate limit errors

    OpenRouter published a guide on handling 429 Too Many Requests errors from OpenAI, Anthropic and its own API, testing three fixes on a 60-prompt batch. The OpenAI Python SDK's default retries left 23 of 60 requests failed in 12 seconds, while a retry that waits for Retry-After or X-RateLimit-Reset completed all 60 in 89 to 92 seconds in Python and 106 to 218 seconds in TypeScript. A models fallback finished all 60 in 13 seconds for $0.000066, and with provider failover on, all 150 pinned-provider requests succeeded where up to 32 failed with fallbacks off.

  12. 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.

  14. LangChain Blog58

    LangChain explains retrieval techniques for AI agent context engineering

    LangChain published a blog post on how retrieval design shapes context engineering for AI agents, using a competitive analysis agent built on its Deep Agents harness as the example. The post covers three failure modes, inefficient searching, loading too much context at once, and irrelevant or stale results, and pairs each with techniques such as glob and grep filesystem tools, paging through large files with read_file offsets, offloading oversized tool results to files, and delegating reading to subagents under a retrieve, offload, delegate pattern. It also warns that retrieved text is data rather than instructions and can carry prompt injection, so outputs should be checked against cited sources.

  15. 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.

  17. Meta Engineering Blog58

    Meta details how Your Algorithm turns natural-language preferences into Facebook Feed and Reels recommendations

    Meta published an engineering overview of Your algorithm, a suite of AI tools that lets Facebook users state in natural language what they want to see in Feed and Reels, with a single editable page listing Interested In and Not Interested In topics. Each change is validated by a smaller LLM, enriched with embeddings and search keywords, and saved to a shared store that both Feed and Reels read, while a fine-tuned LLM reranker removes content matching newly added Not Interested In topics. Meta says a supplementary tail load and client-side cache invalidation make changes take effect faster, and that a separate LLM judge grades outputs such as the Lately Summary before launch.

  18. Meta Engineering Blog58

    Meta explains TLX, a hardware-native GPU compiler extension to Triton for production ML

    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.

  19. 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.

  20. Mastra Blog38

    Mastra Forward: Mastra launches AI engineering service with forward-deployed engineers and agent factory

    Mastra launched Mastra Forward, an AI engineering service that pairs a named forward-deployed engineer with Mastra Factory, an agent swarm installed in the customer's repo. The annual engagement costs less than one senior AI hire and includes an Enterprise license with hosted, self-hosted, or BYOC deployment plus a Trace Intelligence pool. Mastra says its factory produces 750+ PRs per week.