GPT-6.1 Terra beats Claude Haiku 4.5 on cost and accuracy in tests

An October 28, 2026 test on a Linux server compared GPT-6.1 Terra in Codex CLI with Claude Haiku 4.5 in Claude Code across three tasks. Terra completed 9 of 9 hard-kit runs versus 3 of 9 for Haiku 4.5, with a median round costing $0.0689 versus $0.2127 at API Standard rates.

• Easy kit: both 9 of 9, rounds cost $0.0630 for Terra and $0.1393 for Haiku 4.5
• Hard kit: 9 of 9 fully right for Terra versus 3 of 9 for Haiku 4.5
• Median round times were 31.5 seconds for Terra and 105.7 seconds for Haiku 4.5
• Both list $1 per million input and $5 per million output tokens