A few weeks ago I used CursorBench 3.1 to pick a Claude Code model by budget and target use.
The leaderboard is now on CursorBench 3.2, Opus 5 is out, and it rearranges the whole picture: Fable 5 has essentially no reason to exist anymore.
I’ll start from the conclusions, and data / graphs will follow.
Four cost/score bands
| Band | Target score | Winner | Cost | Why |
|---|---|---|---|---|
| Budget | ≤48% | Sonnet 5 (Low) | $1.30 | Cheapest point on the chart |
| Everyday | ≤63% | Opus 5 (Low) | $2.55 | Beats Opus 4.8 Max and Sonnet 5 Max at under half their cost |
| Mid | 64–67% | Opus 5 (Medium–High) | $3.29–$3.91 | Edges out Fable 5 High for 45% of the cost |
| High | 68–70% | Opus 5 (Extra High–Max) | $7.35–$8.23 | Fable 5 needs $11.73–$17.32 to get there |
Past 70.0%, only Fable 5 Max gets higher — by 0.5 points, for $9.09 more per task.
Cost vs. score graph
Opus 5
Fable 5
Opus 4.8
Sonnet 5
Tokens vs. score graph
Opus 5
Fable 5
Opus 4.8
Sonnet 5
Takeaways
- Opus 5 Low is the new floor for anything serious: 62.8% at $2.55 and 18.5k output tokens. That’s above Opus 4.8 Max (62.3%, $5.77) and Sonnet 5 Max (61.5%, $6.45).
- Opus 4.8 has no reason to be picked anymore. Its ceiling is beaten by the cheapest Opus 5 tier. Only its Low tier undercuts Opus 5 on price, and $0.53 of savings for 9.7 fewer points is a bad trade.
- Fable 5 is dominated tier by tier: Low loses to Opus 5 Low, High loses to Opus 5 High at 45% of the cost, Extra High loses to Opus 5 Extra High. Only Fable 5 Max stays on top — 70.5% vs 70.0% — and that half-point costs $9.09 extra and 1.7× the output tokens. Noise, at a premium.
- Sonnet 5 keeps exactly one useful point: Low, at $1.30. Its Medium tier (52.4%, $2.16) is now worse and pricier than Opus 4.8 Low, and $0.39 short of Opus 5 Low, which scores 10 points higher.
- Opus 5’s cliff is between High and Extra High: +2.6 points for 1.9× the cost and 1.9× the tokens. High is where the value stops.
In short:
- Throwaway edits → Sonnet 5 Low.
- Everyday work → Opus 5 Low/Medium.
- Feature work, gnarly debugging → Opus 5 High. This is the sweet spot.
- Need the ceiling? → Opus 5 Extra High. Fable 5 isn’t worth reaching for anymore.
Data pulled from the CursorBench 3.2 leaderboard (updated 2026-07-08); the live page may have since updated with more models or revised numbers.
Raw data
| Model | Tier | Score % | Cost/task | Tokens/task |
|---|---|---|---|---|
| Opus 5 | Max | 70.0 | $8.23 | 61,838 |
| Opus 5 | Extra High | 69.3 | $7.35 | 54,239 |
| Opus 5 | High | 66.7 | $3.91 | 27,932 |
| Opus 5 | Medium | 64.3 | $3.29 | 23,612 |
| Opus 5 | Low | 62.8 | $2.55 | 18,529 |
| Fable 5 | Max | 70.5 | $17.32 | 103,525 |
| Fable 5 | Extra High | 68.4 | $11.73 | 64,971 |
| Fable 5 | High | 66.5 | $8.77 | 43,747 |
| Fable 5 | Medium | 65.2 | $6.80 | 30,366 |
| Fable 5 | Low | 62.1 | $4.46 | 18,182 |
| Opus 4.8 | Max | 62.3 | $5.77 | 71,411 |
| Opus 4.8 | Extra High | 59.4 | $4.50 | 51,121 |
| Opus 4.8 | High | 58.0 | $3.15 | 33,548 |
| Opus 4.8 | Medium | 56.1 | $2.81 | 28,384 |
| Opus 4.8 | Low | 53.1 | $2.02 | 19,624 |
| Sonnet 5 | Max | 61.5 | $6.45 | 92,882 |
| Sonnet 5 | Extra High | 58.7 | $4.16 | 52,871 |
| Sonnet 5 | High | 56.9 | $3.19 | 39,483 |
| Sonnet 5 | Medium | 52.4 | $2.16 | 26,200 |
| Sonnet 5 | Low | 47.7 | $1.30 | 16,269 |