OpenCode

AI for software development — on your premises.

State-of-the-art coding models deployed locally. Not a single line of your code leaves your infrastructure.

Model tiers

Choose the model tier that matches your hardware and performance requirements.

OpenCode model tiers with SWE-bench scores
Tier	Model	Params	VRAM	SWE-bench
Pro	MiniMax M2.5	230B	115 GB	74.0%
Air	Qwen 3.5	122B	100 GB	72.0%
Flash	Devstral Small 2	30B	24 GB	68.0%

SWE-bench Verified scores. Models updated as part of the annual retainer.

In-depth analysis of pull requests, diff reviews, and targeted bug fixes with context-aware suggestions.

Spec-driven development from requirements to working code. Build complete features with multi-file context.

Connect directly to your repositories. Review PRs, understand codebases, and push changes — all locally.

Index and query large codebases. Ask questions, trace call paths, and understand unfamiliar projects quickly.

Generate comprehensive test suites, identify edge cases, and ensure coverage across your codebase.

Delegate complex tasks to specialised sub-agents. Parallelise work across multiple AI instances.

Book a discovery session to understand your hardware requirements and integration approach.