Cline Agents Harness Unlocks Benchmark-Leading Performance from NVIDIA Nemotron 3.5 Super VL

NVIDIA Nemotron 3.5 Super VL, an open-weights model, was evaluated within the Cline agent framework on Terminal-Bench 2.1 and in an internal Slack-based support agent used for task triage, bug investigation, and issue resolution. With thinking mode disabled, Nemotron 3.5 Super VL 120B A12B scored 61.8% in Cline, compared with 53.9%, 51.7%, and 44.9% across three unnamed open-source harnesses, although the comparison does not isolate which Cline features caused the difference or quantify compute savings. Cline’s approach includes prompting the model to plan before coding, detecting stalled attempts through structured failure events, and using targeted verification checks, with examples involving chess move analysis and terminal signal handling. The internal agent can create sandboxes and containers and escalate complex tasks to more advanced models; in one cited case, it investigated a Windows diagnostics-export failure caused by an existing Downloads folder and an EEXIST error. The team presents the benchmark and production-style use cases as complementary ways to assess the model while continuing to refine the harness based on observed failures and daily engineering work.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.