Claude Opus 5.5 vs Gemini 4 Carbon

Every finance benchmark where both have a score (Anthropic and Google). Data refreshed 2 Nov 2026.

Shared benchmarks
4
Claude Opus 5.5 only
1
Gemini 4 Carbon only
1

Shared benchmarks, side by side

Each benchmark is drawn on its own scale. Bars are not comparable across rows.

  • Claude Opus 5.5
  • Gemini 4 Carbon
  • Excel Modeling Benchmark (EMB) accuracy %, higher is better

    Claude Opus 5.5:75.8
    Gemini 4 Carbon:76.5 (better)
  • Finance Agent Benchmark Partial Credit accuracy %, higher is better

    Claude Opus 5.5:67.8 (better)
    Gemini 4 Carbon:67.2
  • Vals Finance Index index accuracy %, higher is better

    Claude Opus 5.5:65.6
    Gemini 4 Carbon:66.3 (better)
  • Tax Agent Bench accuracy % (all-pass x citation factor), higher is better

    Claude Opus 5.5:45.6
    Gemini 4 Carbon:48.9 (better)
Benchmark
Claude Opus 5.5
Gemini 4 Carbon

6 benchmarks, 4 scored for both. Bold marks the better score where both have one.

Excel Modeling Benchmark (EMB)accuracy %, higher is better
75.8
Vals.ai leaderboard (Official board, 29 Oct 2026)
76.5
Vals.ai leaderboard (Official board, 29 Oct 2026)
Finance Agent BenchmarkPartial Credit accuracy %, higher is better
67.8
Vals.ai leaderboard (Official board, 29 Oct 2026)
67.2
Vals.ai leaderboard (Official board, 29 Oct 2026)
Vals Finance Indexindex accuracy %, higher is better
65.6
Vals.ai leaderboard (Official board, 29 Oct 2026)
66.3
Vals.ai leaderboard (Official board, 29 Oct 2026)
Tax Agent Benchaccuracy % (all-pass x citation factor), higher is better
45.6
Vals.ai leaderboard (Official board, 29 Oct 2026)
48.9
Vals.ai leaderboard (Official board, 29 Oct 2026)
APEX-AccountingMean Score (% criteria passed), higher is better
Not scored
70.7
Mercor APEX-Accounting leaderboard (Official board, 1 Nov 2026)
FinanceBench% correct, no filing given (150-question sample), higher is better
60.3
Moneda run, 2026-11-02 (Run by Moneda, 2 Nov 2026)
Not scored

Scores from different benchmarks use different scales and are not added up. Settings can differ between sources; see each leaderboard for details.