# Claude Opus 5 vs GPT-5.4

> Claude Opus 5 is the stronger model overall, scoring 67.7 to 59.5 on the Noometry Index. GPT-5.4 costs 1.8× less per token, which makes it the better buy when Claude Opus 5's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/claude-opus-5-vs-gpt-5-4
- Last updated: 2026-10-10
- Shared benchmarks: 49

## Summary

- They share 49 benchmarks with published results for both. Claude Opus 5 scores higher in 9 categories and GPT-5.4 in 1 category; 10 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where Claude Opus 5 leads 77.1 to 61.7.
- The biggest single-benchmark swing is ProofBench: 99% for Claude Opus 5 and 56% for GPT-5.4.
- GPT-5.4 is cheaper at $2.50 / $15 per million input/output tokens, against $5 / $25 for Claude Opus 5.
- GPT-5.4 accepts more context: 1.05M tokens versus 1M.

## Snapshot

| | Claude Opus 5 | GPT-5.4 |
|---|---|---|
| Provider | Anthropic | OpenAI |
| Noometry Index | 67.7 | 59.5 |
| Rank | 4 | 14 |
| Context | 1M | 1.05M |
| Input $/M | $5 | $2.50 |
| Output $/M | $25 | $15 |
| Weights | Proprietary | Proprietary |

## Coding

- Claude Opus 5: 67.0 (#5)
- GPT-5.4: 52.8 (#28)

| Benchmark | Claude Opus 5 | GPT-5.4 |
|---|---|---|
| DeepSWE | 73.6% | 51.8% |
| LMArena WebDev | 1694 | 1465 |
| SciCode | 56.4% | 56.6% |
| WeirdML | 91.8% | 77.7% |
| LMArena Coding | 1533 | 1497 |
| ALE-Bench | 2,165 | 1,607 |
| SWE-bench Verified | — | 76.9% |
| FrontierCode | 53.4% | — |
| CursorBench | 46.6% | — |
| FrontierSWE | 52% | — |
| GSO | — | 31.4% |
| MirrorCode | — | 15.6% |
| AlgoTune | — | 1.85 |

## Agentic & Tool Use

- Claude Opus 5: 55.5 (#1)
- GPT-5.4: 46.5 (#12)

| Benchmark | Claude Opus 5 | GPT-5.4 |
|---|---|---|
| APEX-Agents | 65.8% | 52.4% |
| τ²-bench Banking | 48.7% | 39.4% |
| PostTrainBench | 35% | 19% |
| GBAEval | 79.6% | 45.1% |
| Vending-Bench 2 | 11,182 | 6,144 |
| Terminal-Bench | — | 81.8% |
| OSWorld 2.0 | 31.4% | — |
| DeepResearch Bench | — | 35.1% |
| BALROG | 63.4% | — |
| GDP.pdf | 24% | — |
| LMArena Search | — | 1197 |
| METR Time Horizons | — | 74.3% |

## Reasoning

- Claude Opus 5: 77.1 (#3)
- GPT-5.4: 61.7 (#17)

| Benchmark | Claude Opus 5 | GPT-5.4 |
|---|---|---|
| ARC-AGI-2 | 90.4% | 74% |
| NYT Connections (extended) | 94.3% | 91.3% |
| ARC-AGI-1 | 97.5% | 93.7% |
| CritPt | 29.1% | 23.4% |
| Chess Puzzles | 42% | 44% |
| EBR-Bench | 45.7% | 25.4% |
| LMArena Hard Prompts | 1525 | 1485 |
| Mystery Game Puzzles | 59% | 37% |
| DTBench | 97.9% | 94.4% |
| LMCA | 64.5% | 52% |
| Epoch Capabilities Index | 162.67 | 156.92 |
| SimpleBench | 80.6% | — |
| Kagi LLM Benchmark | — | 63.8% |
| EnigmaEval | — | 16% |
| Thematic Generalization | — | 80% |
| Bench to the Future 3 | 0.12 | — |
| ForecastBench | — | 59.5 |

## Math

- Claude Opus 5: 86.1 (#6)
- GPT-5.4: 73.3 (#14)

| Benchmark | Claude Opus 5 | GPT-5.4 |
|---|---|---|
| FrontierMath (Tiers 1-3) | 85.6% | 78.6% |
| FrontierMath Tier 4 | 73.2% | 49% |
| OTIS Mock AIME 2024-2025 | 98.9% | 97.8% |
| ProofBench | 99% | 56% |
| LMArena Math | 1533 | 1488 |
| MathArena Final-Answer Competitions | — | 83.1% |
| FrontierMath (Feb 2025 set) | — | 47.6% |
| FrontierMath Tier 4 (v1) | — | 27.1% |

## Knowledge

- Claude Opus 5: 66.8 (#8)
- GPT-5.4: 65.3 (#13)

| Benchmark | Claude Opus 5 | GPT-5.4 |
|---|---|---|
| GPQA Diamond | 93.9% | 93.3% |
| SimpleQA Verified | 59.9% | 45.1% |
| LMArena Expert | 1558 | 1507 |
| Humanity's Last Exam | — | 36.2% |
| Vectara Hallucination Rate | — | 7% |

## Multimodal

- Claude Opus 5: 51.1 (#6)
- GPT-5.4: 43.8 (#19)

| Benchmark | Claude Opus 5 | GPT-5.4 |
|---|---|---|
| LMArena Vision | 1321 | 1302 |
| Blueprint-Bench 2 | 30.4% | 27.1% |
| Furniture Assembly | 60.8% | 37.5% |
| LMArena Document | 1516 | 1471 |

## Multilingual

- Claude Opus 5: 58.8 (#3)
- GPT-5.4: 56.4 (#22)

| Benchmark | Claude Opus 5 | GPT-5.4 |
|---|---|---|
| LMArena Non-English | 1500 | 1465 |
| LMArena Chinese | 1575 | 1520 |
| LMArena French | 1522 | 1493 |
| LMArena German | 1522 | 1472 |
| LMArena Japanese | 1510 | 1484 |
| LMArena Korean | 1526 | 1449 |
| LMArena Russian | 1507 | 1480 |
| LMArena Spanish | 1519 | 1453 |

## Instruction Following

- Claude Opus 5: 79.3 (#5)
- GPT-5.4: 77.3 (#25)

| Benchmark | Claude Opus 5 | GPT-5.4 |
|---|---|---|
| LMArena Instruction Following | 1516 | 1469 |

## Long Context

- Claude Opus 5: 46.5 (#20)
- GPT-5.4: 50.3 (#8)

| Benchmark | Claude Opus 5 | GPT-5.4 |
|---|---|---|
| LMArena Longer Query | 1516 | 1474 |
| CL-bench | — | 27.9% |
| CL-bench Life | — | 21.7% |

## Writing & Preference

- Claude Opus 5: 79.3 (#2)
- GPT-5.4: 72.1 (#18)

| Benchmark | Claude Opus 5 | GPT-5.4 |
|---|---|---|
| LMArena Text | 1506 | 1469 |
| LMArena Creative Writing | 1491 | 1439 |
| EQ-Bench Creative Writing | 2133 | 1840 |
| EQ-Bench 4 | 1385 | 1272 |
| LMArena Multi-Turn | 1496 | 1482 |

## FAQ

### Is Claude Opus 5 better than GPT-5.4?

Claude Opus 5 is the stronger model overall, scoring 67.7 to 59.5 on the Noometry Index. GPT-5.4 costs 1.8× less per token, which makes it the better buy when Claude Opus 5's lead doesn't matter for your workload.

### Which is cheaper, Claude Opus 5 or GPT-5.4?

GPT-5.4 is cheaper. It lists at $2.50 per million input tokens and $15 per million output tokens; Claude Opus 5 lists at $5 and $25.

### Is Claude Opus 5 or GPT-5.4 better for coding?

Claude Opus 5 scores higher on coding benchmarks: 67.0 versus 52.8 in the Noometry coding category.

### Which has the bigger context window?

GPT-5.4 does, with 1.05M tokens against 1M.

### How many benchmarks do Claude Opus 5 and GPT-5.4 share?

49 benchmarks have published results for both models. Claude Opus 5 has 57 scored results on Noometry and GPT-5.4 has 68.