GPT-6.1 Terra in Codex: low to high effort stays within 7% on cost

GPT-6.1 Terra was tested in Codex CLI across four reasoning effort levels on two task sets. Low, medium and high stayed within 7% of each other in cost per round, while xhigh cost 1.07x medium on the easy kit and 1.49x on the hard kit. Every level solved the hard kit fully; the only miss was one low-effort run on the easy kit.

• Low, medium and high differ by no more than 7% in cost per round on both kits
• Xhigh costs 1.07x medium on the easy kit and 1.49x on the hard kit
• Hard kit: all 9 runs fully right at every level, including 8 hidden tests
• Low missed 1 of 9 easy-kit runs: 7,799 mm² instead of 8,999 mm²