ForgeCode: Top open source coding agent in Terminal-Bench 2.0 – Tensorlake (www.tensorlake.ai)

ForgeCode has emerged as the top open-source coding agent in the Terminal-Bench 2.0 benchmark, showcasing significant improvements in model performance. By utilizing the Gemini 3.1 Pro model with a new orchestration layer, ForgeCode achieved an impressive 80.2% performance score, marking a 25 percentage point increase from previous benchmarks without altering the underlying model. This advancement underscores its innovative approach to tool call schemas and parallel execution, which enhances efficiency and reduces formatting errors. Key to ForgeCode's performance is its use of flattened schemas, allowing for clearer and more consistent tool calls, resulting in fewer errors. Additionally, its parallel execution model, facilitated by the join_all() function, enables simultaneous tool calls rather than sequential execution, significantly speeding up tasks by 3–5 times. Although ForgeCode's architectural design allows for multiple agent instances to operate concurrently—streamlining complex tasks—it currently faces limitations, such as a lack of persistent memory and a smaller user community compared to others in the field. These technical advancements position ForgeCode as a valuable tool for the AI/ML community, pushing the boundaries of coding agent capabilities.