Together AI's New CLI Lets You Swap Proprietary Models for Open Alternatives Inside Your IDE

Developer tools are about to get cheaper. Together AI just released Together Link, a free, MIT-licensed command-line tool that intercepts AI model requests from your code editor and redirects them to open-source alternatives. In practical terms: you can keep using Claude Code or Codex the same way you always have, but swap in models like Kimi K3 or GLM 5.3 instead of paying for proprietary APIs.

Setup is minimal. One install command gets it running, and the tool includes an Auto router that automatically selects which model to use for each coding session based on task complexity and performance. This removes the friction of manually switching between models or managing multiple API keys.

The pitch is straightforward economics. Together AI claims users can save over 50% on AI inference costs by routing through open models, which typically charge per token at a fraction of what OpenAI or Anthropic ask. For teams running dozens of developers with frequent model calls, that compounds quickly. The tool works with Claude Code, Codex, OpenCode, Pi, and both the Claude and ChatGPT desktop apps, covering most of the popular development environments where engineers already spend their time.

This represents a subtle but meaningful shift in how developers think about AI tooling. For months, the industry narrative has centered on closed, proprietary models as the default. Together Link is betting that developers care more about flexibility and cost than being locked into a single provider. By making it trivial to swap models without changing your IDE or workflow, it lowers the barrier to experimenting with open alternatives.

The open model landscape has matured significantly. Models like Kimi K3 and GLM 5.3 are now competitive enough for many coding tasks. But adoption has been hampered by friction. Developers would need to set up new environments, learn new tools, or juggle multiple interfaces. Together Link removes that friction layer.

The timing is interesting. As AI API costs become a material line item for engineering teams and developer free tiers tighten, cost arbitrage between closed and open models becomes a real decision point. A tool that makes that arbitrage painless could find real traction among cost-conscious teams.