Amazon Bedrock now supports open models for AI coding agents via integration with OpenCode

Amazon Bedrock now supports open models (NVIDIA Nemotron 3 Super 120B, Moonshot AI Kimi K3, and GPT-OSS 120B from OpenAI) as a foundation for AI coding agents, through integration with OpenCode – an open source terminal coding agent written in Go, which according to the vendor connects to more than 75 language model providers including Amazon Bedrock. AWS describes on its blog how to set up OpenCode with open models on Bedrock and how to choose a model depending on the type of task: Kimi K3 for debugging and architecture analysis thanks to a 1M token context, GPT-OSS 120B for multi-file code generation, and Nemotron 3 Super 120B for high-volume generation.

According to AWS, with this setup the code, prompts, and responses remain within the customer's AWS account, the models inherit the same security (IAM, CloudTrail, PrivateLink) as closed models, and Bedrock falls within the scope of compliance programs such as HIPAA, SOC 2, ISO 27001, FedRAMP, and GDPR. AWS states there are three pricing tiers – Priority for latency-sensitive production operation, Standard for pay-per-token, and Flex with a price 50% lower for workloads with variable latency – and adds that the global cross-region profile works out roughly 10% cheaper than the geographic profile. The default limits are, according to AWS, 100 million tokens per minute and 10 thousand requests per minute.

In the text, AWS cites a McKinsey study according to which 76% of organizations plan to increase their use of open source AI, and leading companies in AI adoption are 40% more likely to be users of open models, as well as a 2026 Gartner analysis according to which agentic workflows multiply token consumption 5 to 30 times. As an example of performance, AWS cites the company CrowdStrike, whose fine-tuned Nemotron model is said, according to this source, to have achieved 96% accuracy of valid queries compared to 61% for GPT-4o and 94% for the Claude Sonnet 4.5 model – this is a claim taken from the AWS blog, not an independently verified result. According to the article, the company Ethara.AI is deploying the described architecture in production for multi-agent orchestration of development and research workflows.

The source text was only partially available, and the description of the architecture and specific configuration steps is not complete. See the source article for details.