Together AI releases Together Link to cut developer costs

Together AI has introduced Together Link, a free, MIT-licensed command-line interface currently in beta for macOS and Linux. The tool connects existing developer coding agents directly to open-source models hosted on Together AI's platform. Supported agents include Claude Code, Claude Desktop, Codex, ChatGPT Desktop, OpenCode, and Pi. By substituting expensive proprietary models with open alternatives, the company claims developers can achieve overall cost savings of more than 50 percent, and between 50 and 80 percent compared to sessions running entirely on Anthropic's Opus 5.5.

The CLI operates without a local proxy or daemon, routing queries directly to Together's hosted gateway. An automated router evaluates the first task of a session to allocate resources. If a developer has an Anthropic API key, the system routes tasks between Opus 5.5 and GLM 5.3. Without one, it routes between GLM 5.3 and GLM 5.3 Flash. Within Claude Code, the tool maps tiers to specific open models with 1M context windows: Opus runs Kimi K3, Fable runs GLM 5.3, Sonnet runs GLM 5.3 Flash, and Haiku runs DeepSeek V4.1 Flash. Users can also pin a specific model using flags, such as targeting GLM 5.3 with the main flag.

Pricing for these hosted models is highly competitive. Kimi K3 costs $3.00 per million input tokens and $15.00 per million output tokens, while GLM 5.3 is priced at $1.40 input and $4.40 output. DeepSeek V4.1 Flash and MiniMax M3 cost $0.30 input and $1.20 output. Together AI notes that as of September 30, 2026, it serves the largest OpenRouter token share for DeepSeek V4.1 Flash at 40.8 percent, GLM 5.3 Flash at 28.2 percent, and Kimi K3 at 23.1 percent.

For practitioners, installation requires a single curl command that automatically installs Bun if necessary. Sessions are billed to an existing Together API key, and the terminal prints token and dollar totals upon exit. Claude Code even displays estimated spend alongside equivalent Opus costs, and developers can track their gateway expenses over a seven-day period using a simple usage command. This setup allows engineers to maintain their preferred workflows and agent interfaces while drastically reducing their monthly API bills.