According to AWS Machine Learning, open weight models are changing the economics of building and deploying AI at scale. Rapid advances in intelligence and efficiency enable companies to match each workload with the right balance of capability, speed and cost. AWS is building a future where organizations can adapt open scale innovations with the reliability and security they need for production. Kimi K3 from Moonshot AI is available today on the Amazon Bedrock platform, providing a powerful new option for coding and knowledge work. According to information from Moonshot AI, Kimi K3 is the company's most powerful model and the first open model to reach 2.8 trillion parameters. It combines native rendering capabilities with a 1 million token context window and provides an approximate 2.5x improvement over Kimi K2 in terms of scalability. These advances make Kimi K3 well suited for long-running coding and knowledge workflows that require persistent context across large repositories, documents and images. Kimi K3 is the first open weight model on Amazon Bedrock that
I support explicit hint caching, which helps reduce latency and input costs when reusing context between model calls. The launch of Kimi K3 reflects AWS's ongoing investment in open weight models on Amazon Bedrock. Since 2025, Bedrock has added dozens of open weight models from providers including DeepSeek, Google, MiniMax, Mistral AI, Moonshot AI, NVIDIA, OpenAI and Qwen. Support for this expanding sample continues through the continued development of inference technology that serves these models at scale. In 2026, Bedrock added support for tool calls, structured inference, reasoning, response streaming, and an API for chat responses and terminations. Because these are platform capabilities and not model-by-model integrations, new open weight models can take advantage of them as they become available on Amazon Bedrock. Like all open weight models on Amazon Bedrock, you can use the Kimi K3 without compromising your safety. Your data is processed within AWS data boundaries and is not transferred to the model provider or used to train the underlying model. Zero data retention is always enabled for queries
in output, while null operator access prevents even AWS operators from accessing your hints and completions during output. These precautions allow you to use open scale models with confidence while maintaining control of your data. To get started with Kimi K3 on Amazon Bedrock, open your Amazon Bedrock console, go to Test > Site, and select Kimi K3 as your model. You can also programmatically invoke the model using the bedrock-runtime endpoint, which supports OpenAI-compatible APIs for chat replies and ends, as well as the Amazon Bedrock Invoke and Converse APIs. You can call Kimi K3 via cross-region output profile. For workloads without regional restrictions, we recommend using the global profile global.moonshotai.kimi-k3, which forwards every request to any supported commercial AWS region around the world. Global inference costs between regions are approximately 10% lower than the geographic profile. The US geographic profile, us.moonshotai.kimi-k3, maintains processing within the US to comply with data storage requirements. Requirements: Active AWS account
with access to Amazon Bedrock. Python 3.10+. AWS Identity and Access Management (AWS IAM) permissions for model invocation are bedrock:InvokeModel, bedrock:InvokeModelWithResponseStream, and bedrock:CreateInference. Here's a quick example that uses the OpenAI SDK and the aws-bedrock-token-generator library for Python to generate short-lived bearer tokens for authentication with Amazon Bedrock. Optimize output with explicit hint caching: Long-running coding and knowledge workflows often resubmit stable context, such as repository instructions, tool definitions, or reference documents. With explicit hint caching, you can define reusable hint prefixes so that subsequent requests can use the cached content. When a request matches a cached prefix, Amazon Bedrock can reduce response latency and input token costs. Caching for Kimi K3 on Amazon Bedrock: You can mark the exact end of a reusable prompt prefix (after at least 1,024 tokens) by adding prompt_cache_breakpoint to the supported input content. In explicit mode, tokens written to the cache will be charged
at a higher rate, but are then cached for at least 30 minutes. For subsequent requests that are cached, input tokens will be charged at a reduced rate and will not count towards input token per minute quotas. You can explore the Moonshot AI sample repository on AWS for more examples. Kimi K3 in practice: In addition to using the API directly, you can use Kimi K3 through a wide range of coding assistants, personal agents, and agent frameworks that support Amazon Bedrock in particular or OpenAI-compatible models in general. There are several popular coding agents available to developers today, so let's look at OpenCode as an example. OpenCode is open source, model independent and has a native amazon-bedrock model provider that uses the Converse API. To start, you can configure the amazon-bedrock provider either in your opencode.json configuration files at the user level or at the project level, as shown in the OpenCode documentation. Once the provider is configured, OpenCode will automatically detect available Amazon Bedrock models, which
You can select them using the /models command. For example, a minimal ~/.config/opencode.json file might look like this: { "$schema": "https://opencode.ai/config.json", "model": "amazon-bedrock/global.moonshotai.kimi-k3", "provider": { "amazon-bedrock": { "options": { "region": "us-west-2", "profile": "PLACEHOLDER-YOUR-AWS-PROFILE-NAME" } } } }. After setting up the Amazon Bedrock provider, you can use the /models command to switch the models to global.moonshotai.kimi-k3 and start developing. Kimi K3 can create significant features and work on long-term tasks. In the next video we try to create a browser-based one-page game to get started. Productivity Agents: Beyond coding, Hermes Agent is an example of an open assistant for overall productivity. It can be used through the desktop application or popular instant messengers, as well as through the terminal, and supports use cases such as deep research and task automation, where the Kimi K3 can also work well. As stated in their documentation, Hermes natively supports models on Amazon Bedrock. To get started, run the hermes model from your terminal. Scroll through the list
distributors to "AWS Bedrock" (Hermes mislabels "Amazon Bedrock" as "AWS Bedrock"). If prompted, select the source AWS region to which you want to send Hermes requests. Select either the default credential chain (recommended) to use the AWS Command Line Interface credentials already configured in your environment, or create an Amazon Bedrock API key. Select Kimi K3 from the auto-detected list of models, or if it is not available, enter global.moonshotai.kimi-k3 as the custom model name. If you are using named profiles to manage multiple AWS credentials in your environment, then at the time of writing you will need to set the AWS_PROFILE environment variable or use your default Hermes profile. Otherwise, you can switch to an API key. Keep an eye on the open issue here for updates on support for installing an AWS profile via the Hermes configuration file. Once the Amazon Bedrock provider is set up and the model is configured, you can start using Kimi K3 for your agent-based workflows in Hermes. For example, watch the following short video in which we ask a development agent
There is a personalized curriculum. Kimi K3 is available today on Amazon Bedrock via US and global geographic profiles for cross-region output. See the Bedrock documentation for a complete list of supported regions. For pricing information, see Amazon Bedrock Pricing. Try Kimi K3 in the Amazon Bedrock console or explore the Moonshot AI sample repository on AWS on GitHub. The material was prepared by a neural network based on the original publication. AI may make mistakes in details (numbers, names, dates, quotes) - for important information, check with the original source. Migration of multi-model AI agents to the Amazon Bedrock platform Security researchers used Claude to hack OpenAI Kyutai released a Voice of Reason model for verbally solving math problems. Nokia opens access to AnyJev, a library for decision-making without training. OpenAI introduced GPT-6 Sol and Luna with new API pricing, reducing the cost by 50%.
