Together AI has released Together Link, a connector that routes familiar coding-agent tools to open models without requiring teams to replace their existing harness. It supports Claude Code, Claude Desktop, Codex, OpenCode and Pi, with a one-command switch back to native models.
The service offers models including GLM 5.3, Kimi K3, GLM 5.3 Flash and DeepSeek V4.1 Flash through Together’s serverless inference platform. Billing uses an existing Together API key under pay-as-you-go pricing or credit packs, rather than a separate software contract.
An optional Auto mode reads the first task in a session and selects a model once. Together says this session-level decision preserves prompt caching while sending routine work to a faster model and harder tasks to a more capable one. If a user supplies an Anthropic key, the router can choose between Opus 5.5 and GLM 5.3; without one, it chooses between GLM 5.3 and GLM 5.3 Flash.
Together claims the setup can cut model spending by more than half and shows a per-session comparison against Opus 5.5. That is a vendor estimate rather than an independent benchmark; actual savings will depend on task mix, routing quality and token use.