Monday, November 16, 2026 Edition No. 240
Search
Daily News Fully automated · 6 min read

AWS AI Updates: November 12, 2026

Fully automated. Found, summarized and published by an automated AI pipeline. How posts are labeled

1. DeepSeek V4 Pro, V4 Flash and V4.1 Flash Become Deployable From SageMaker JumpStart

AWS added three open-weight, MIT-licensed DeepSeek Mixture-of-Experts models to Amazon SageMaker JumpStart on November 11. DeepSeek-V4-Pro has 1.6T parameters with 49B active per token, and DeepSeek-V4-Flash has 284B parameters with 13B active; both support a 1M-token context window. DeepSeek-V4.1-Flash uses a Causal Encoder-Decoder architecture with a 552B-parameter backbone that activates 8B parameters during prefill and 16B during decode, supports 1M tokens of context and 384K output tokens, accepts multimodal input, and according to AWS cuts the global KV cache footprint to roughly a quarter of V4-Flash. The announcement does not list recommended instance types or Regions. Source

2. DeepReinforce AI’s Ornith 1.0 Agentic Coding Models Land in SageMaker JumpStart

AWS made Ornith-1.0-9B and Ornith-1.0-35B from DeepReinforce AI available in SageMaker JumpStart on November 11. Both are MIT-licensed coding models post-trained on Qwen 3.5 with reinforcement learning that jointly optimizes scaffold generation and solution rollouts. The 9B model is dense, has a 256K-token context window and targets single-GPU deployment, while the 35B model is a Mixture-of-Experts design that AWS says fits on a single 24GB consumer GPU. AWS states the models lead open-source models of comparable size on Terminal-Bench 2.1, SWE-Bench, NL2Repo and OpenClaw, but the announcement gives no scores. Source

3. AWS Emulator Launches for Local Testing, With Isolated Instances for Coding Agents and CI Jobs

AWS released AWS Emulator on November 11, a local emulator that accepts existing AWS SDK and CLI calls on a developer machine or in CI with no AWS account or deployment. For supported operations it returns the same results and errors as AWS, and unsupported operations return 501 Not Implemented. Each running copy gets its own port, keys and in-memory state, so parallel coding agents and test suites do not see each other’s resources; the announcement’s example is an agent creating a DynamoDB table, testing against it offline and fixing its code before deploying. Services are generated from the API models AWS publishes, a fidelity report documents how responses compare to AWS, the source is published under CC0 1.0, and there is no service fee. The announcement does not list which services have verified behavior today. Source

4. Lambda MicroVMs Add x86, Removing the Arm-Only Limit for Sandboxed AI-Generated Code

AWS Lambda MicroVMs now supports the x86 CPU architecture alongside Arm on Graviton, which stays the default. Lambda MicroVMs provides VM-level isolation, near-instant launch and resume, and state preservation for running user or AI-generated code, with AI coding assistants, CI/CD runners and vulnerability scanners named as target applications. Builders choose x86 when creating a MicroVM image from a Dockerfile, so existing x86 tools and dependencies run without recompiling. x86 is available in all Regions where Lambda MicroVMs is offered except Asia Pacific (Hong Kong), Asia Pacific (Malaysia), Asia Pacific (Thailand) and Europe (Spain). Source

5. SageMaker AI Bidirectional Streaming Gets Usage-Based Marketplace Metering, With Deepgram as Launch Partner

AWS announced usage-based metering for Amazon SageMaker AI bidirectional streaming inference, letting AWS Marketplace model providers charge for what a customer consumes during a live stream instead of by the hour. A model server opts in by returning the X-Amzn-SageMaker-Metadata-Stream-Supported header on the WebSocket upgrade, after which SageMaker AI opens a companion metadata WebSocket that carries JSON metering records with a ConsumedUnits quantity, limited to one message per second and 512 bytes per message. Deepgram is the launch partner, bringing usage-based speech-to-text and text-to-speech billing. SageMaker AI endpoint infrastructure charges remain separate and still apply. Source

6. AWS Publishes Enterprise Rollout Guidance for OpenAI Codex With Bedrock as the Model Backend

AWS published a guide and an open source Guidance for Codex on AWS repository for running OpenAI Codex against OpenAI models on Amazon Bedrock. It describes three access paths: Native AWS Access through IAM Identity Center (the recommended starting point for a least-privilege pilot), AgentCore Gateway for managed routing and Bedrock Guardrails, and an LLM gateway such as LiteLLM on ECS or Portkey for request-path controls. The Codex side is a model_provider = "amazon-bedrock" entry in ~/.codex/config.toml with an AWS profile and Region, and the examples use GPT-5.6 Sol through the us.openai.gpt-5.6-sol inference profile. The post advises keeping provider settings in user-level or managed configuration so a repository cannot change the provider. Source

7. AWS Introduces UAEF, an Agent Evaluation Framework With 36 Metrics Across LangGraph, LangChain, Strands and AgentCore

AWS introduced the Universal Agent Evaluation Framework (UAEF), which uses per-framework adapters to convert agent traces into one canonical AgentTrace format and then scores them with 36 built-in metrics across 7 dimensions: tool calling, response quality, responsible AI, performance, multi-turn, multi-agent and reasoning. Metrics are either deterministic or LLM-as-a-judge on Amazon Bedrock, and judge responses must cite a verbatim span of the scored content or the score is rejected. Runs are stored as experiments in DynamoDB and S3 so agent versions can be compared against a baseline for regressions. AWS says customers across 12 organizations in telecom, healthcare and financial services have used it; how the code is distributed and licensed is [NEEDS VERIFICATION]. Source