GPT-6.1 Sol Tops KingBench 4 Team Score

The result is less a solo-model coronation than a strong case for routing architecture and delegation through Opus while using Sol for execution.

• –The 44.2/60 score measures a coordinated model-and-harness system, not GPT-6.1 Sol in isolation.
• –Sol’s $2 input and $10 output pricing, plus $0.10 cached input, make repeated agentic coding workflows substantially cheaper.
• –KingBench-style tests reward shipping functional software under hard time limits, exposing whether agents complete integrations rather than merely producing polished interfaces.
• –Developers should evaluate merge conflicts, review burden, latency, and cache efficiency alongside benchmark scores.
• –Community tests suggest Sol can produce larger diffs and slower responses, so lower inference cost may shift some expense into human review.