OpenCode Tops Cross-Model Harness Benchmark

OpenCode’s biggest advantage may be its portability: the harness, not just the underlying model, materially shapes agent performance.

• –The study compares OpenCode, Codex CLI, Pi CLI, and DeepAgents across six benchmarks and four core models.
• –OpenCode leads on cross-model fit, making it especially compelling for teams switching between proprietary, open-weight, and local models.
• –Results vary by model and benchmark, so the paper does not establish OpenCode as the universal best harness.
• –Valid completion rate and task success sometimes diverge, highlighting reliability as a separate engineering metric.
• –Developers should benchmark their own model-harness combinations instead of relying on headline scores.