Together Link: Frontier-Quality Open Models in the Harness You Already Use
Together Link integrates frontier open models like GLM 5.3 and Kimi K3 into existing coding agents, reducing costs by over 50%. This solution supports tools such as Claude Code and Codex, allowing teams to switch between premium and open models seamlessly without altering workflows or requiring new logins.
Tags
Quick summary
Together Link integrates frontier open models like GLM 5.3 and Kimi K3 into existing coding agents, reducing costs by over 50%. This solution supports tools such as Claude Code and Codex, allowing teams to switch between premium and open models seamlessly without altering workflows or requiring new logins.
Together Link: Frontier-Quality Open Models in the Harness You Already Use
Engineering teams are increasingly relying on coding agents to automate tasks ranging from one-line fixes to full-scale rewrites. While these tools have become commonplace, the cost of running them on premium, closed models has escalated rapidly. At scale, organizations are spending anywhere from tens of thousands to millions of dollars a month on proprietary infrastructure. Together AI addresses this challenge with Together Link, a solution designed to integrate frontier-quality open models directly into the agent harnesses teams already use.
According to Together AI, this integration cuts model spend by over 50% without requiring changes to existing workflows. The service connects familiar tools to the best open models available on the Together AI platform, allowing teams to maintain their current settings and logins while accessing a fraction of the bill.
Seamless Integration with Existing Workflows
The primary value proposition of Together Link is its ability to operate within the environment developers already know. Together AI states that the solution supports a wide array of existing coding agents, including Claude Code, Claude Desktop, Codex (available in the ChatGPT app and CLI), OpenCode, and Pi.
When a team deploys Together Link, the agent opens exactly as before. There is no need to learn new interfaces or manage separate contracts. The transition is designed to be frictionless: users simply create an account to obtain a Together AI API key, and the agent connects to the new infrastructure. If a team wishes to revert to native closed models, the switch takes only one command. This approach ensures that productivity is not disrupted while cost structures are optimized.
Intelligent Model Routing
A critical component of the Together Link architecture is the Together AI Router. This system reads the first task of each session to determine the appropriate model for the workload. The routing logic is designed to balance cost efficiency with capability:
- Quick Fixes: Simple tasks are routed to fast, low-cost models.
- Hard Problems: Complex challenges are directed toward frontier-capable models.
Together AI specifies that if a user provides an Anthropic key, the router facilitates a choice between Opus 5.5 and GLM 5.3. In the absence of an Anthropic key, the system routes between GLM 5.3 and GLM 5.3 Flash. This routing decision occurs once per session, which ensures that prompt caching continues to function effectively throughout the interaction.
The models highlighted as leaders in this ecosystem include Kimi K3 and GLM 5.3, which Together AI notes now handle many of the hardest coding tasks at a fraction of the per-token price compared to closed labs. For everyday coding tasks, smaller and capable models like GLM 5.3 Flash and DeepSeek V4.1 Flash are utilized.
Cost Transparency and Savings
Together Link provides immediate visibility into cost savings. A per-session tracker displays what a user spent during a specific session alongside a projection of what that same session would have cost on Opus 5.5. This comparison allows engineering organizations to quantify the financial impact of switching to open models in real time.
Billing is managed through the existing Together API key, utilizing serverless pay-as-you-go or credit packs. Together AI emphasizes that users do not need to negotiate a separate contract to access these capabilities.
Infrastructure and Performance
Together Link runs on Together's serverless inference platform. Together AI asserts that this is the same infrastructure developers already choose for these models on OpenRouter. The company reports serving the largest share of OpenRouter tokens for several key models, citing specific metrics as of September 30, 2026:
- DeepSeek V4.1 Flash: 40.8% of tokens served.
- GLM 5.3 Flash: 28.2% of tokens served.
- Kimi K3: 23.1% of tokens served.
Together AI claims competitive speed and pricing across the board for these models. The serverless nature of the platform aligns with the preferences of developers who prioritize flexibility and scalability.
Getting Started
Deployment of Together Link is designed to be rapid. Together AI indicates that teams can be up and running on open models in just a few minutes. The setup process involves a single command to install the link, followed by the creation of an API key.
The installation command provided by Together AI is:
curl -fsSL https://link.together.ai/install | bashOnce installed, the user creates an account to generate the necessary API key. The system then connects the agent to Together AI, enabling the intelligent routing and cost-saving features described above.
Conclusion
Together Link represents a strategic shift in how engineering organizations approach AI infrastructure. By bridging the gap between the high performance of frontier models and the cost efficiency of open models, it offers a practical solution to the rising expenses of coding agents. With support for major tools like Claude Code and Codex, and an intelligent routing system that optimizes model selection per session, Together Link allows teams to retain their existing workflows while significantly reducing operational costs. As Together AI notes, the result is the same agent, the same workflow, and a fraction of the bill.
