Independent comparison
AI Model Comparison: 18 Models, One Index, Real Bills
The short answer
As of November 12, 2026, Claude Fable 5.2 has the highest Intelligence Index of the 18 models we track (65), at $26.54 per evaluation task. If you are paying the bill, GPT-6.1 Terra is the value pick: 51 points — 78% of Fable's score — for $0.96 per task, 3.6% of the cost. If your input is audio or video, Gemini 4 Carbon is the only model with a published cost figure that accepts both (60, $7.76).
| Model | Index | Cost / task | Why | |
|---|---|---|---|---|
| Best overall | Claude Fable 5.2 | 65 | $26.54 | Highest score in the table |
| Best value | GPT-6.1 Terra | 51 | $0.96 | 78% of the top score for 3.6% of the cost |
| Audio + video | Gemini 4 Carbon | 60 | $7.76 | Only priced model that takes speech and video |
What makes this table different
Most comparison pages rank models by one number and stop. This one prices the bill.
- Cost per task, not just token price. Two models on the same $4/$20 rate card can land at $7.76 and $9.95 per task — the rate card cannot show that, the bill does.
- One index, one version, one reasoning tier. Mixing LLM Stats scores with Artificial Analysis prices, or a max-effort score with a low-effort price, produces a table that is internally inconsistent. Every row here is the same index at the same version at the model's highest reasoning setting.
- Gaps are labelled, never filled. When the source publishes nothing for a cell, the cell says so.
N/Ais not zero, andnot computableis not a value.
Head-to-head comparisons
- Claude Opus 5.5 vs GPT-6.1 Terra — 59 vs 51, but 10.5× the cost per task.
- Gemini 4 Carbon vs Gemini 4 Argon — identical $4/$20 list price, $7.76 vs $9.95 per task.
- Kimi K3.1 Pro vs Claude Sonnet 5.5 — 9 index points for 2.7× the price.
- Muse Wonder Full vs Gemini 4 Carbon — both multimodal; only one has a cost figure.
How to read the numbers
The Intelligence Index is a composite score, not a percentage. The useful comparison is between rows, not the absolute number. Three columns decide most bills and appear in no rate card: cost per task, time per task, and output length. The full table, the method and the update policy are on the ranking page; the sourcing rules are in methodology.