Picking the "right" Claude Code model

Today I stumbled upon a LinkedIn post about their CursorBench 3.1, which I found very interesting. With Sonnet 5 out and Fable 5 available again, I used the benchmark to compare all three families by budget and target use.
I’ll start from the conclusions, and data / graphs will follow.

Three cost/score bands

The data splits cleanly into three regimes depending on the score you actually need:

Band Target score Winner Cost Why
Budget ≤55% Sonnet 5 (Low–Medium) $1.46–$2.57 Cheapest points on the chart
Mid 55–58% Opus 4.8 (High) $4.41 Matches Sonnet 5 Extra High’s score ($5.23) for less
High 60%+ Fable 5 (Low) $5.70 Beats every Opus 4.8/Sonnet 5 tier above 60%, on cost and score

Past ~65%, only Fable 5 gets there at all.

Cost vs. score graph

Fable 5: Low 64.2% at $5.70, Medium 69.8% at $8.27, High 70.6% at $10.81, Extra High 72.0% at $13.74, Max 72.9% at $18.02. Opus 4.8: Low 54.3% at $2.93, Medium 56.6% at $3.83, High 58.4% at $4.41, Extra High 62.1% at $6.14, Max 63.8% at $7.59. Sonnet 5: Low 47.7% at $1.46, Medium 54.9% at $2.57, High 57.0% at $3.74, Extra High 58.4% at $5.23, Max 61.2% at $6.87.
Fable 5 Opus 4.8 Sonnet 5

Tokens vs. score graph

Fable 5: Low 64.2% at 18,882 tokens, Medium 69.8% at 28,507, High 70.6% at 37,173, Extra High 72.0% at 48,754, Max 72.9% at 63,842. Opus 4.8: Low 54.3% at 22,726, Medium 56.6% at 31,684, High 58.4% at 36,788, Extra High 62.1% at 55,622, Max 63.8% at 77,370. Sonnet 5: Low 47.7% at 17,028, Medium 54.9% at 27,469, High 57.0% at 41,735, Extra High 58.4% at 58,228, Max 61.2% at 93,485.
Fable 5 Opus 4.8 Sonnet 5

Takeaways

  • Sonnet 5 Medium and Opus 4.8 Low are roughly equivalent on cost and tokens. Sonnet’s slight score edge makes it the pick for everyday work.
  • Sonnet 5 High edges out Opus 4.8 Medium, but Opus 4.8 High wins that same score bracket outright. You may never want to use Sonnet 5 Extra High.
  • Need something more than Opus 4.8 High? Just switch to Fable 5 Low. Considering the output tokens you may also prefer it to Opus 4.8 Max.

In short:

  • Quick, low-stakes edits → Sonnet 5 Low/Medium.
  • Everyday feature work → Opus 4.8 High.
  • Need to go smarter? Fable 5 Low is the deal.

Data pulled from a screenshot of the CursorBench 3.1 leaderboard; the live page may have since updated with more models or revised numbers.

Raw data

Model Tier Score % Cost/task Tokens/task
Fable 5 Max 72.9 $18.02 63,842
Fable 5 Extra High 72.0 $13.74 48,754
Fable 5 High 70.6 $10.81 37,173
Fable 5 Medium 69.8 $8.27 28,507
Fable 5 Low 64.2 $5.70 18,882
Opus 4.8 Max 63.8 $7.59 77,370
Opus 4.8 Extra High 62.1 $6.14 55,622
Opus 4.8 High 58.4 $4.41 36,788
Opus 4.8 Medium 56.6 $3.83 31,684
Opus 4.8 Low 54.3 $2.93 22,726
Sonnet 5 Max 61.2 $6.87 93,485
Sonnet 5 Extra High 58.4 $5.23 58,228
Sonnet 5 High 57.0 $3.74 41,735
Sonnet 5 Medium 54.9 $2.57 27,469
Sonnet 5 Low 47.7 $1.46 17,028
Share: X (Twitter) Facebook LinkedIn