Together AI has launched Together Code Interpreter (TCI), an API that executes LLM-generated Python code in a secure sandboxed environment and returns results (stdout/stderr) for use in agentic workflows. Sessions last 60 minutes, support multiple executions, and are priced at $0.03/session. TCI also supports MCP, enabling integration with tools like Cursor and Windsurf. A key use case is reinforcement learning: Agentica used TCI to run 1,024 concurrent unit tests per batch during RL training, helping produce DeepCoder-14B-Preview, which achieved 60.6% Pass@1 on LiveCodeBench matching o3-mini performance at 14B parameters.
