Claude Opus 5 vs Gemini 4 Carbon
Gemini 4 Carbon leads Claude Opus 5 on coding, 86.6 to 81.7 on the Vibe Coding Index, and Gemini 4 Carbon costs 2.5x less per output token ($10.00 vs $25.00 per million). Updated November 14, 2026; 62 benchmarks have a result for both.
vs boards won of 62 shared
Claude Opus 5 or Gemini 4 Carbon: which to pick
Pick Gemini 4 Carbon for coding: it scores 86.6 on the Vibe Coding Index against 81.7 for Claude Opus 5, though on 4 benchmarks against 5.
Pick Claude Opus 5 for work measured by Design Arena: agents building mobile apps or Design Arena: spreadsheets, where it leads.
Model AClaude Opus 5Anthropic81.7/1005 independent benchmarks$1.33 per solved taskModel BGemini 4 CarbonGoogle86.6/1004 independent benchmarksNo cost per solved task
- Shared results
- 62
- Claude Opus 5 only
- 83
- Gemini 4 Carbon only
- 11
- Like-for-like categories
- 8 of 9
Only benchmarks that tested both models count toward a pick; a result only one model has is shown but not compared.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, six public coding benchmarks
Gemini 4 Carbon
Gemini 4 Carbon scores 86.6 against 81.7.
5 and 4 benchmarks behind the scores
Intelligence
AA Intelligence Index
Gemini 4 Carbon
Gemini 4 Carbon leads, 60.0 against 44.2.
Same board, both measured
Agentic work
AA Agentic Index
No call
Only Claude Opus 5 has a AA Agentic Index result, 56.5.
Directional only
Reasoning
GPQA Diamond
No call
Only Claude Opus 5 has a GPQA Diamond result, 93.4%.
Directional only
Knowledge
MMLU Pro
No call
Only Claude Opus 5 has a MMLU Pro result, 91.6%.
Directional only
Web design
Design Arena: websites
Gemini 4 Carbon
Gemini 4 Carbon leads, 1319 against 1304.
Same board, both measured
General
Vals Index
No call
Only Gemini 4 Carbon has a Vals Index result, 70.5%.
Directional only
Long documents
Listed context window
Gemini 4 Carbon
Gemini 4 Carbon takes 1.05M tokens against 1M.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
Gemini 4 Carbon
Claude Opus 5 $0.018, Gemini 4 Carbon $0.0070 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only Claude Opus 5 has the number, $1.33.
Directional only
At a glance
- Coding index
- Claude Opus 5 81.7 from 5 benchmarks, Gemini 4 Carbon 86.6 from 4 benchmarks.
- Widest category
- Reasoning, 13.9 points apart on average over 12 shared benchmarks.
- Widest single gap
- EBR-bench: Claude Opus 5 45.7% against Gemini 4 Carbon 87.6%, 41.9 points apart on a 0 to 100 scale.
- Output price
- Claude Opus 5 $25.00 against Gemini 4 Carbon $10.00 per million output tokens, 2.5x apart.
- Context window
- Claude Opus 5 takes 1M tokens, Gemini 4 Carbon 1.05M.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | Claude Opus 5 | Bars | Gemini 4 Carbon | Ties |
|---|---|---|---|---|
| Coding | 0 | 12 | ||
| Intelligence | 0 | 2 | ||
| Agentic work | 2 | 4 | ||
| Reasoning | 0 | 12 | ||
| Math | 1 | 4 | ||
| Knowledge | 1 | 5 | ||
| Web design | 1 | 4 | ||
| General | 1 | 13 | ||
| All shared boards | 6 | 56 |
CodingLike-for-likeSix of these boards feed the Vibe Coding Index012
| Benchmark | Claude Opus 5 | Gemini 4 Carbon | Leader |
|---|---|---|---|
| LiveBench coding | Not measured | 91.0% | Only Gemini 4 Carbon measured |
| LMArena WebDev | 1691 | 1768 | Gemini 4 Carbon |
| AA Coding Index | 78.0 | Not measured | Only Claude Opus 5 measured |
| LiveCodeBench (Vals AI) | 89.0% | Not measured | Only Claude Opus 5 measured |
| SWE-bench Verified (Vals AI) | 97.0% | Not measured | Only Claude Opus 5 measured |
| Terminal-Bench 4 (Vals AI) | 53.5% | 63.1% | Gemini 4 Carbon |
| Vibe Code Bench (Vals AI) | 88.4% | 89.1% | Gemini 4 Carbon |
| Code migration (Vals AI) | 57.5% | Not measured | Only Claude Opus 5 measured |
| ProgramBench (Vals AI) | 3.0% | Not measured | Only Claude Opus 5 measured |
| IOI (Vals AI) | 84.3% | Not measured | Only Claude Opus 5 measured |
| Terminal-Bench 2.1 (Vals AI) | 84.6% | Not measured | Only Claude Opus 5 measured |
| Vibe Code Bench 1-100 (Vals AI) | 28.5% | Not measured | Only Claude Opus 5 measured |
| SkillsBench (Vals AI) | 60.4% | Not measured | Only Claude Opus 5 measured |
| CyberBench (Vals AI) | 65.4% | Not measured | Only Claude Opus 5 measured |
| SciCode | 56.4% | Not measured | Only Claude Opus 5 measured |
| Terminal-Bench 4.0 (Artificial Analysis run) | 49.0% | 65.2% | Gemini 4 Carbon |
| FrontierCode 2.0 | 31.2% | 40.6% | Gemini 4 Carbon |
| FrontierCode | 53.4% | Not measured | Only Claude Opus 5 measured |
| FrontierSWE | 52.0% | 61.2% | Gemini 4 Carbon |
| WeirdML | 91.8% | Not measured | Only Claude Opus 5 measured |
| MirrorCode | 87.8% | Not measured | Only Claude Opus 5 measured |
| ALE-Bench | 2164.6 | 2374.7 | Gemini 4 Carbon |
| LMArena Text: coding | 1532 | 1564 | Gemini 4 Carbon |
| LMArena Text: software and IT | 1527 | 1552 | Gemini 4 Carbon |
| LMArena WebDev: React | 1713 | 1792 | Gemini 4 Carbon |
| LMArena WebDev: HTML | 1679 | 1770 | Gemini 4 Carbon |
| LMArena WebDev: from an image | 1664 | Not measured | Only Claude Opus 5 measured |
| SciCode | 13.8% | Not measured | Only Claude Opus 5 measured |
| SWE-Bench Pro V2 | 99.4% | Not measured | Only Claude Opus 5 measured |
| SWE Atlas: Codebase QnA | 63.2% | Not measured | Only Claude Opus 5 measured |
| SWE Atlas: Test Writing | 62.2% | Not measured | Only Claude Opus 5 measured |
| GameBench 2 | 84.8 | Not measured | Only Claude Opus 5 measured |
| Terminal-Bench 3.0 | 42.7% | Not measured | Only Claude Opus 5 measured |
| Terminal-Bench 4.0 | 53.9% | Not measured | Only Claude Opus 5 measured |
| CursorBench 4.0 | 46.6% | 60.4% | Gemini 4 Carbon |
| ProgramBench | 4.5% | Not measured | Only Claude Opus 5 measured |
| Neither model has a result on the other 16 boards here | |||
IntelligenceLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | Claude Opus 5 | Gemini 4 Carbon | Leader |
|---|---|---|---|
| AA Intelligence Index | 44.2 | 60.0 | Gemini 4 Carbon |
| Epoch Capabilities Index | 164.7 | 169.8 | Gemini 4 Carbon |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index24
| Benchmark | Claude Opus 5 | Gemini 4 Carbon | Leader |
|---|---|---|---|
| AA Agentic Index | 56.5 | Not measured | Only Claude Opus 5 measured |
| Design Arena: agents building web apps | 1244 | Not measured | Only Claude Opus 5 measured |
| Design Arena: agents building full-stack apps | 1282 | Not measured | Only Claude Opus 5 measured |
| Design Arena: agents building mobile apps | 1348 | 1284 | Claude Opus 5 |
| Design Arena: agents building Android apps | 1245 | Not measured | Only Claude Opus 5 measured |
| Design Arena: agents building Godot games | 1257 | Not measured | Only Claude Opus 5 measured |
| Design Arena: agents building games | 1199 | Not measured | Only Claude Opus 5 measured |
| Design Arena: PowerPoint slides | 1225 | Not measured | Only Claude Opus 5 measured |
| Design Arena: spreadsheets | 1229 | 1224 | Claude Opus 5 |
| LiveBench research | Not measured | 92.5% | Only Gemini 4 Carbon measured |
| LiveBench browser | Not measured | 80.0% | Only Gemini 4 Carbon measured |
| LiveBench long agentic | Not measured | 72.7% | Only Gemini 4 Carbon measured |
| SREBench (Vals AI) | 12.2% | Not measured | Only Claude Opus 5 measured |
| Finance Agent (Vals AI) | 58.6% | Not measured | Only Claude Opus 5 measured |
| Tax Agent Bench (Vals AI) | 47.1% | 51.8% | Gemini 4 Carbon |
| Legal Research Bench (Vals AI) | 55.3% | 55.8% | Gemini 4 Carbon |
| Harvey's Legal Agent Benchmark (Vals AI) | Not measured | 17.5% | Only Gemini 4 Carbon measured |
| Excel Modeling Benchmark (Vals AI) | 73.6% | Not measured | Only Claude Opus 5 measured |
| BioMysteryBench (Vals AI) | 79.3% | Not measured | Only Claude Opus 5 measured |
| Terminal-Bench-Science (Vals AI) | 27.1% | Not measured | Only Claude Opus 5 measured |
| Vals Science Index | 47.9% | Not measured | Only Claude Opus 5 measured |
| Time Horizon Index (Vals AI) | 18.8% | Not measured | Only Claude Opus 5 measured |
| CUA-bench (Vals AI) | 9.0% | Not measured | Only Claude Opus 5 measured |
| APEX-Agents | 65.8% | 84.2% | Gemini 4 Carbon |
| OSWorld 2.0 | 31.4% | Not measured | Only Claude Opus 5 measured |
| BALROG | 63.4% | Not measured | Only Claude Opus 5 measured |
| PostTrainBench | 35.7% | Not measured | Only Claude Opus 5 measured |
| Vending-Bench 2 | $11,182 | $16,484 | Gemini 4 Carbon |
| LMArena Search | 1239 | Not measured | Only Claude Opus 5 measured |
| MCP Atlas | 85.8% | Not measured | Only Claude Opus 5 measured |
| τ³-Banking | 48.7% | Not measured | Only Claude Opus 5 measured |
| Neither model has a result on the other 13 boards here | |||
ReasoningLike-for-likePublic boards, shown as published, kept out of the index012
| Benchmark | Claude Opus 5 | Gemini 4 Carbon | Leader |
|---|---|---|---|
| GPQA Diamond | 93.4% | Not measured | Only Claude Opus 5 measured |
| LiveBench reasoning | Not measured | 90.6% | Only Gemini 4 Carbon measured |
| MysteryMechanism (Vals AI) | 37.4% | Not measured | Only Claude Opus 5 measured |
| MMMU Pro (Vals AI) | 89.9% | Not measured | Only Claude Opus 5 measured |
| GPQA Diamond (Artificial Analysis run) | 93.7% | Not measured | Only Claude Opus 5 measured |
| Humanity's Last Exam (Artificial Analysis run) | 54.9% | Not measured | Only Claude Opus 5 measured |
| HLE-Diamond (Artificial Analysis run) | 44.8% | 69.1% | Gemini 4 Carbon |
| CritPt v2, hard subset (Artificial Analysis run) | 50.0% | 69.0% | Gemini 4 Carbon |
| GPQA Diamond (Epoch AI run) | 93.9% | 96.5% | Gemini 4 Carbon |
| ARC-AGI | 97.5% | 98.5% | Gemini 4 Carbon |
| ARC-AGI-2 | 90.4% | 94.2% | Gemini 4 Carbon |
| SimpleBench | 80.6% | 92.6% | Gemini 4 Carbon |
| Chess puzzles | 43.0% | 78.0% | Gemini 4 Carbon |
| Mystery game puzzles | 59.0% | 73.0% | Gemini 4 Carbon |
| EBR-bench | 45.7% | 87.6% | Gemini 4 Carbon |
| LMArena Text: hard prompts | 1525 | 1551 | Gemini 4 Carbon |
| LMArena Text: expert | 1552 | 1582 | Gemini 4 Carbon |
| LMArena Text: hard prompts in English | 1522 | 1545 | Gemini 4 Carbon |
| LMArena Vision: diagrams | 1345 | Not measured | Only Claude Opus 5 measured |
| LMArena Vision: homework | 1335 | Not measured | Only Claude Opus 5 measured |
| Humanity's Last Exam (Scale run) | 38.6% | Not measured | Only Claude Opus 5 measured |
| GoBench | 2255 | Not measured | Only Claude Opus 5 measured |
| Neither model has a result on the other 4 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index14
| Benchmark | Claude Opus 5 | Gemini 4 Carbon | Leader |
|---|---|---|---|
| OTIS Mock AIME | 98.9% | 100.0% | Gemini 4 Carbon |
| FrontierMath, tiers 1 to 3 | 85.6% | 90.5% | Gemini 4 Carbon |
| FrontierMath, tier 4 | 73.2% | 95.1% | Gemini 4 Carbon |
| ProofBench | 99.0% | Not measured | Only Claude Opus 5 measured |
| LMArena Text: math | 1532 | Not measured | Only Claude Opus 5 measured |
| LMArena Text: mathematical work | 1543 | 1538 | Claude Opus 5 |
| MathArena | 63.6% | 85.4% | Gemini 4 Carbon |
| Neither model has a result on the other 7 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index15
| Benchmark | Claude Opus 5 | Gemini 4 Carbon | Leader |
|---|---|---|---|
| MMLU Pro | 91.6% | Not measured | Only Claude Opus 5 measured |
| LegalBench (Vals AI) | 87.0% | Not measured | Only Claude Opus 5 measured |
| CorpFin (Vals AI) | 73.2% | Not measured | Only Claude Opus 5 measured |
| TaxEval (Vals AI) | 75.1% | Not measured | Only Claude Opus 5 measured |
| MortgageTax (Vals AI) | 72.1% | Not measured | Only Claude Opus 5 measured |
| MedCode (Vals AI) | 63.6% | Not measured | Only Claude Opus 5 measured |
| AA-Omniscience | 37.6 | 53.5 | Gemini 4 Carbon |
| SimpleQA Verified | 59.9% | 77.7% | Gemini 4 Carbon |
| LMArena Text: science | 1533 | 1535 | Gemini 4 Carbon |
| LMArena Text: medicine and healthcare | 1527 | 1525 | Claude Opus 5 |
| LMArena Text: legal and government | 1528 | 1542 | Gemini 4 Carbon |
| LMArena Text: business and finance | 1499 | 1520 | Gemini 4 Carbon |
| FACTS Benchmark Suite | Not measured | 78.4% | Only Gemini 4 Carbon measured |
| PRBench Legal | Not measured | 53.3% | Only Gemini 4 Carbon measured |
| Neither model has a result on the other 7 boards here | |||
Web designLike-for-likePublic boards, shown as published, kept out of the index14
| Benchmark | Claude Opus 5 | Gemini 4 Carbon | Leader |
|---|---|---|---|
| Design Arena: websites | 1304 | 1319 | Gemini 4 Carbon |
| Design Arena: UI components | 1341 | 1357 | Gemini 4 Carbon |
| Design Arena: game dev | 1330 | 1326 | Claude Opus 5 |
| Design Arena: data viz | 1326 | Not measured | Only Claude Opus 5 measured |
| Design Arena: 3D | 1321 | 1434 | Gemini 4 Carbon |
| Design Arena: all code categories | 1317 | 1348 | Gemini 4 Carbon |
| Design Arena: SVG | 1306 | Not measured | Only Claude Opus 5 measured |
| Design Arena: ASCII art | 1353 | Not measured | Only Claude Opus 5 measured |
| LiveBench design | Not measured | 80.2% | Only Gemini 4 Carbon measured |
GeneralLike-for-likePublic boards, shown as published, kept out of the index113
| Benchmark | Claude Opus 5 | Gemini 4 Carbon | Leader |
|---|---|---|---|
| Vals Index | Not measured | 70.5% | Only Gemini 4 Carbon measured |
| LiveBench data analysis | Not measured | 86.5% | Only Gemini 4 Carbon measured |
| Public Benefits Bench (Vals AI) | 76.9% | Not measured | Only Claude Opus 5 measured |
| MedScribe (Vals AI) | 91.0% | Not measured | Only Claude Opus 5 measured |
| Vals Multimodal Index | 73.9% | Not measured | Only Claude Opus 5 measured |
| Long context reasoning (AA-LCR) | 82.0% | 79.3% | Claude Opus 5 |
| GDPval-AA | 1729 | 1733 | Gemini 4 Carbon |
| DTBench | 97.9% | Not measured | Only Claude Opus 5 measured |
| LMCA | 64.5% | Not measured | Only Claude Opus 5 measured |
| Furniture assembly | 60.8% | 77.5% | Gemini 4 Carbon |
| LMArena Text | 1507 | 1538 | Gemini 4 Carbon |
| LMArena Text: instruction following | 1518 | 1539 | Gemini 4 Carbon |
| LMArena Text: creative writing | 1494 | 1533 | Gemini 4 Carbon |
| LMArena Vision | 1317 | Not measured | Only Claude Opus 5 measured |
| LMArena Text: English | 1507 | 1530 | Gemini 4 Carbon |
| LMArena Text: non-English | 1500 | 1534 | Gemini 4 Carbon |
| LMArena Text: Chinese | 1575 | 1613 | Gemini 4 Carbon |
| LMArena Text: German | 1520 | Not measured | Only Claude Opus 5 measured |
| LMArena Text: Japanese | 1521 | Not measured | Only Claude Opus 5 measured |
| LMArena Text: Korean | 1504 | Not measured | Only Claude Opus 5 measured |
| LMArena Text: French | 1520 | Not measured | Only Claude Opus 5 measured |
| LMArena Text: Spanish | 1522 | Not measured | Only Claude Opus 5 measured |
| LMArena Text: Russian | 1507 | 1560 | Gemini 4 Carbon |
| LMArena Text: Polish | 1522 | Not measured | Only Claude Opus 5 measured |
| LMArena Text: multi-turn | 1499 | 1569 | Gemini 4 Carbon |
| LMArena Text: longer queries | 1516 | 1549 | Gemini 4 Carbon |
| LMArena Text: writing and language | 1499 | 1542 | Gemini 4 Carbon |
| LMArena Text: entertainment and media | 1480 | 1527 | Gemini 4 Carbon |
| LMArena Vision: English | 1310 | Not measured | Only Claude Opus 5 measured |
| LMArena Vision: Chinese | 1402 | Not measured | Only Claude Opus 5 measured |
| LMArena Vision: reading text | 1325 | Not measured | Only Claude Opus 5 measured |
| LMArena Vision: humor | 1294 | Not measured | Only Claude Opus 5 measured |
| LMArena Vision: creative writing | 1309 | Not measured | Only Claude Opus 5 measured |
| LMArena Document | 1509 | Not measured | Only Claude Opus 5 measured |
| EQ-Bench 4 | 1365 | Not measured | Only Claude Opus 5 measured |
| Neither model has a result on the other 14 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | Claude Opus 5 | Gemini 4 Carbon | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
Claude Opus 5$0.018Fits in one request
Gemini 4 Carbon$0.0070Fits in one request
Gemini 4 Carbon costs less
Repository review
50K input + 3K output tokens
Claude Opus 5$0.325Fits in one request
Gemini 4 Carbon$0.130Fits in one request
Gemini 4 Carbon costs less
Agent loop
200K input + 10K output tokens
Claude Opus 5$1.25Fits in one request
Gemini 4 Carbon$0.500Fits in one request
Gemini 4 Carbon costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | Anthropic Claude Opus 5 | Google Gemini 4 Carbon |
|---|---|---|
| Maker | Anthropic | |
| Weights | No public weights listed | No public weights listed |
| Released | July 24, 2026 | November 10, 2026 |
| Context window | 1M | 1.05M |
| Input $/Mtok | $5.00 | $2.00 |
| Output $/Mtok | $25.00 | $10.00 |
| Coding index | 81.7 | 86.6 |
| Independent measurements behind it | 5 | 4 |
| Cost per solved task | $1.33 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is Claude Opus 5 better than Gemini 4 Carbon for coding?
Pick Gemini 4 Carbon for coding: it scores 86.6 on the Vibe Coding Index against 81.7 for Claude Opus 5, though on 4 benchmarks against 5. Pick Claude Opus 5 for work measured by Design Arena: agents building mobile apps or Design Arena: spreadsheets, where it leads.
Which is cheaper, Claude Opus 5 or Gemini 4 Carbon?
Claude Opus 5 costs $5.00 per million input tokens and $25.00 per million output tokens; Gemini 4 Carbon costs $2.00 and $10.00. Cost per solved task is known for Claude Opus 5 only.
Which has the bigger context window, Claude Opus 5 or Gemini 4 Carbon?
Gemini 4 Carbon, with 1.05M tokens against 1M.
How many benchmarks is the Claude Opus 5 vs Gemini 4 Carbon comparison based on?
62 benchmarks have a result for both models as of November 14, 2026. Claude Opus 5's coding index rests on 5 independent benchmarks and Gemini 4 Carbon's on 4. Where only one model has a result, the page says so instead of naming a winner.