How the frontier models compare on capability, human preference, tool use, speed and price, kept apart so one score never hides a tradeoff.
| Rank | Model | Overall |
|---|---|---|
| 01 | GPT 5.6 Sol Max | 82% |
| 02 | Claude Fable 5 Max Effort | 81% |
| 03 | Claude Opus 5 xHigh Effort | 80% |
| 04 | GPT 5.5 xHigh | 80% |
| 05 | GPT 5.6 Terra Max | 80% |
| 06 | GPT 5.6 Sol xHigh | 80% |
| 07 | Claude Fable 5 xHigh Effort | 80% |
| 08 | Claude Opus 5 Max Effort | 79% |
| 09 | Claude Opus 5 High Effort | 79% |
| 10 | Claude Opus 4 8 xHigh Effort | 79% |
One number cannot tell you which model to ship.
Capability, price and speed pull against each other.
So they stay on separate boards, with the release each score came from.
All 7 boards are free to read and link to the evidence behind every row. Refresh rates differ by board and are shown on each one.
How the frontier models score when the same tests are run on all of them.
What that capability costs per dollar, per token and per second.