The Pulse
Together AI Brings Open Models to Six Coding Agents With Together Link
Together AI launched Together Link on October 5, a tool that connects open models to coding agents including Claude Code, Codex, OpenCode and Pi. The company says its per-session router and cost tracker can cut spending by more than 50%, bu
AI.info Team · Written with AI ·
Together AI launched Together Link on October 5, offering a way to run open models through several coding agents developers already use. The company says the tool can cut spending by more than 50% while preserving existing logins, settings and workflows.
One interface, several coding agents
Together Link supports Claude Code, Claude Desktop, Codex in the ChatGPT app and command-line interface, OpenCode, and Pi. It connects those tools to Together AI’s hosted models, including Kimi K3, GLM 5.3, GLM 5.3 Flash and DeepSeek V4.1 Flash.
Users install it with a shell command, connect a Together AI API key and launch their coding agent through the service. According to the Together Link site, Claude Code traffic passes through Together’s gateway while users stay signed in, and the Claude and ChatGPT desktop apps open on a separate, reversible profile; the launch post says going back to the native closed models takes one command. The approach is designed to avoid manually rewriting an agent’s existing configuration.
Routing happens once per session
The default “Auto” option sends the first task in a session to a model selected by Together’s router. The company says quick fixes go to lower-cost models, while harder tasks are routed to a more capable option. With an Anthropic key, the router chooses between Opus 5.5 and GLM 5.3; without one, it chooses between GLM 5.3 and GLM 5.3 Flash.
Together says the router makes its choice once at the start of a session, allowing prompt caching to continue working. A receipt at the end shows the session’s spending alongside an estimate of what the same work would have cost on Opus 5.5. Billing runs through the customer’s existing Together API key, using pay-as-you-go inference or credit packs rather than a separate contract.
The savings claim comes with a narrow comparison
Together advertises savings of more than 50%, but its launch post does not explain the workload, token use or calculation behind that figure. The comparison shown in the product is session spending against estimated Opus 5.5 costs, so actual savings will depend on which models and agents a team uses and how much work it sends through them.
The post also gives a snapshot of Together’s share of OpenRouter tokens for three models, dated September 30: 40.8% for DeepSeek V4.1 Flash, 28.2% for GLM 5.3 Flash and 23.1% for Kimi K3. Those figures describe the company’s share of tokens for those specific models, not a market-wide share across coding-agent usage.
Hosted inference, not local model deployment
Together Link runs models on Together AI’s serverless inference platform. Despite the company’s description of the models as open, the launch does not offer local deployment through Link; developers send usage through Together’s service and pay its API rates.
The release positions the product as a way to change the model behind an existing coding agent without changing the agent itself. Its practical test is whether teams can get the advertised cost reduction on their own workloads while keeping their current tools, settings and session habits intact.