Comparing 2 models

Claude Opus 5.5 vs Gemini 4 Carbon

Benchmarks through

Verdict

Claude Opus 5.5 places higher on 13 of the 18 tasks where both are ranked. Gemini 4 Carbon costs 2.0x less per input token. Gemini 4 Carbon accepts 1.0x more context.

  • Choose Claude Opus 5.5 for Math & Reasoning (#1), Frontend & UI (#1), Competition Math (#1).
  • Choose Gemini 4 Carbon for Logic & Problem Solving (#1), a General Assistant (#1), RAG & Long Context (#1).

Built from the published task rankings below; ranks come from each task's own weighted benchmarks, so a win on one task says nothing about another.

metric
Model 1Claude Opus 5.5
Model 2Gemini 4 Carbon
Best for
Competition Math (#1)
a General Assistant (#1)
Provider
anthropic
google
Context
1M
1.0MBest
Input / 1M
$4
$2Best
Output / 1M
$20
$10Best
Tool use
yesBest
yesBest
Vision
yesBest
yesBest
Reasoning
yesBest
yesBest
Open-weight
—
—
Released
Sep 22, 2026
Nov 10, 2026

Measured head to head

Benchmarks both models were run on

EvaluationClaude Opus 5.5Gemini 4 CarbonSource
Design Arena Coding1385 Elo1348 Elosource · 2026-11-15
Design Arena UI1375 Elo1347 Elosource · 2026-11-15
Design Arena Website1326 Elo1319 Elosource · 2026-11-15
FACTS Grounding62.4%77.2%source · 2026-10-24
FrontierMath Tiers 1-3 (Epoch AI)91.2% ± 1.790.5% ± 1.7source · 2026-09-22
GPQA Diamond (Epoch AI)90.6% ± 1.996.5% ± 1.2source · 2026-09-22
LMArena Agent19.517source · 2026-11-11
LMArena Agent Steerability18.619.1source · 2026-11-11
LMArena Agent Task Outcome20.718.4source · 2026-11-11
LMArena Agent Tool Hallucination0.70.5source · 2026-11-11
LMArena Creative Writing1532 Elo1533 Elosource · 2026-11-11
LMArena Instruction Following1535 Elo1539 Elosource · 2026-11-11
LMArena Long Query1528 Elo1549 Elosource · 2026-11-11
LMArena Text1507 Elo1526 Elosource · 2026-11-11
LMArena Text Factuality1505 Elo1521 Elosource · 2026-11-11
LMArena WebDev1815 Elo1768 Elosource · 2026-11-11
OTIS Mock AIME (Epoch AI)100% ± 0100% ± 0source · 2026-09-22
SimpleQA Verified (Epoch AI)72.2% ± 1.477.7% ± 1.3source · 2026-09-22
Terminal-Bench (4.1)65.8%64.5%source · 2026-11-15

Values are as published by each evaluator, in its own units; rows from different evaluators are not on a common scale.