Head-to-head model comparisons

Every comparison uses the same nine axes

Index score, cost per task, minutes per task, input modalities, cost per index point, input and output token price, cache discount and context window. Each cell is tied to a verbatim excerpt of that model's own source page, and a difference is only written when the two cells behind it are both present.

• Claude Opus 5.5 vs GPT-6.1 Terra — the score-versus-bill trade-off, quantified: $10.04 vs $0.96 per task.
• Gemini 4 Carbon vs Gemini 4 Argon — same list price, different bill, 60 vs 50 points.
• Kimi K3.1 Pro vs Claude Sonnet 5.5 — what nine index points cost when the cheaper model is also the slower one.
• Muse Wonder Full vs Gemini 4 Carbon — the multimodal pair where one side has no published cost.

What is not here

A comparison is only published when both models exist on the index at a named reasoning tier. Models announced but not listed, or listed by one aggregator and not confirmable in a second source, do not get a page. Pages are rebuilt weekly: if an index moves by 2 points or more, the numbers and the updated date change together.