Amazon Bedrock Adds Open-Weight Model Support for AI Coding Agents via OpenCode Integration
Summary
- • Amazon Bedrock now supports open-weight AI coding agents — data stays in-account, no per-seat fees
- • OpenCode, a Go-based open-source terminal agent, connects Bedrock to 75+ LLM providers
- • Supported models include Moonshot Kimi K3, OpenAI GPT-OSS 120B, and NVIDIA Nemotron 3 Super 120B
- • 76% of orgs plan to increase open-source AI use as agentic token costs multiply 5–30x (McKinsey)
Details
Amazon Bedrock adds open-weight model support for AI coding agents
Developers can now run AI coding agents using open-weight models on Bedrock, keeping all code and data inside their AWS account with consumption-based pricing rather than per-seat subscriptions
OpenCode: open-source terminal coding agent with Bedrock integration
OpenCode is a Go-built terminal agent that reads and edits files, runs shell commands, and uses Language Server Protocol diagnostics; it connects to 75+ LLM providers including Amazon Bedrock
Supported models: Moonshot Kimi K3, OpenAI GPT-OSS 120B, Nemotron 3 Super 120B
AWS blog demonstrates multi-model workflows on Bedrock routing different coding tasks to the most appropriate open-weight model, enabling model flexibility without provider lock-in
76% of organizations plan to increase open-source AI use (McKinsey 2025)
McKinsey's 2025 report found leading AI adopters are 40% more likely to use open-weight models; Gartner notes agentic workflows multiply token consumption 5–30x, making cost-per-token critical at scale
CrowdStrike fine-tuned Nemotron achieves 96% query accuracy vs GPT-4o's 61%
CrowdStrike's domain-specific fine-tuned NVIDIA Nemotron model outperformed GPT-4o (61%) and Claude Sonnet 4.5 (94%) on security query tasks — evidence that open-weight fine-tuning can surpass proprietary models
Ethara.AI runs Bedrock open-weight multi-agent stack in production
Ethara.AI uses multi-agent orchestration with open-weight models on Bedrock for AI engineering and research workflows at scale, serving as a real-world production reference for the architecture
Amazon Bedrock open-weight coding agent feature details, supported models, and enterprise context
What This Means
Amazon's move to enable open-weight models for coding agents on Bedrock directly addresses the three main enterprise blockers to agentic coding tools: data residency requirements, vendor lock-in, and unpredictable per-seat costs. By pairing Bedrock's managed infrastructure with open-weight models, AWS makes it practical for security-conscious organizations to run capable coding agents without routing proprietary source code through third-party APIs. As Gartner's data shows agentic workflows multiplying token consumption 5–30x over standard usage, cost-per-token is becoming a critical variable — and open-weight models on Bedrock offer a meaningful lever for organizations running at scale.