Together AI has launched Together Link, a tool that integrates existing coding agent harnesses with open-source models to significantly reduce infrastructure costs. The service supports Claude Code, Claude Desktop, Codex, OpenCode, and Pi, allowing engineering teams to switch from expensive closed models like Opus 5.5 to more affordable alternatives without changing their workflow.
- Together Link cuts spending by over 50% by routing tasks to open models such as Kimi K3, GLM 5.3, and DeepSeek V4.1 Flash.
- The tool features an "Auto" router that directs quick fixes to low-cost models and complex problems to frontier capabilities, maintaining prompt caching per session.
- It runs on Together's serverless inference platform, which serves the largest share of tokens for DeepSeek V4.1 Flash (40.8%), GLM 5.3 Flash (28.2%), and Kimi K3 (23.1%) on OpenRouter.
- Setup requires only one command to connect an existing Together AI API key, with billing handled via serverless pay-as-you-go or credit packs.
This integration allows teams to maintain their current tools while leveraging the performance of open models at a fraction of the cost of premium closed-source alternatives.