Daily AI news, sourcedOfficial news, first rumor reports and substantive progress. Each entry keeps the information published at that time.
- Amazon Bedrock adds India in-country inference for Kimi K3
AWS says Amazon Bedrock now offers India-based in-country inference for Moonshot’s Kimi K3, keeping processing in the Mumbai and Hyderabad Regions. The Machine Learning Blog post explains global and geographic cross-Region inference options and shows example code for Indic-language document workloads on Kimi K3.
- Property Finder cuts Bedrock vision-AI cost 86% after fixing evaluation labels inherited from the incumbent model
An AWS Machine Learning blog co-authored with Property Finder says 96.8% of the company’s vision-AI evaluation labels had been inherited from its incumbent model, biasing every comparison. After fresh expert labels and evolutionary prompt optimization on Amazon Bedrock, Property Finder selected Llama 4 Maverick 17B Instruct and reports higher Micro F1 with about 86% lower cost per 1,000 images.
- Amazon Bedrock clarifies batch inference completion behavior and quotas
AWS Bedrock document history for Oct 26 clarifies batch inference completion behavior and quotas on the Capacity and Performance page: on-demand requests are prioritized over batch inference jobs, and job completion is not guaranteed.
- Amazon Bedrock clarifies when Advanced Prompt Optimization returns the original prompt
Amazon Bedrock's documentation history says Advanced Prompt Optimization prompt shortening depends on the evaluation: if the original prompt already scores at the maximum on the evaluation samples, or if the samples do not distinguish between candidate prompts, the original prompt is returned with a compression ratio of 1.0. Guidance on choosing evaluation samples was also updated (doc row dated October 26, 2026).
- Amazon Bedrock Advanced Prompt Optimization adds prompt shortening and accepts application inference profiles
Amazon Bedrock's documentation history says Advanced Prompt Optimization can now shorten prompts, enabled per prompt template, to reduce prompt length while maintaining or improving evaluation scores. Application inference profiles are now accepted as target and judge models, and results include median time to first token and average and median latency.
- Amazon Bedrock adds retrieval instructions to agentic retrieval on managed knowledge bases
AWS says agentic retrieval on fully managed Amazon Bedrock knowledge bases now supports retrieval instructions, which steer how the agent searches knowledge bases and how it answers. Bedrock reviews each instruction before it takes effect, versions can be pinned as immutable, and instructions are applied through the retrievalInstructionConfiguration field on AgenticRetrieveStream, per its Bedrock documentation history (October 19, 2026).
- Amazon Bedrock Guardrails contextual grounding checks now support the Standard tier
Amazon Bedrock Guardrails contextual grounding checks now support the Standard safeguard tier, per the Bedrock documentation history entry dated October 19, 2026. AWS says the Standard tier supports responses that include reasoning or conversational content, returns more consistent scores and supports additional languages.
- Amazon Bedrock adds structured outputs for all Anthropic Claude models
Amazon Bedrock now supports structured outputs for all Anthropic Claude models on both the bedrock-runtime and bedrock-mantle endpoints, where the model is available. Two mechanisms work on both endpoints: a JSON Schema output format that controls the shape of the response, and strict tool use that validates tool parameters against schemas. AWS says structured outputs use constrained decoding to enforce the schema during generation, so every response is valid JSON with required fields present.