# Claude Opus 5 vs GPT-5.2

> Side-by-side benchmark comparison for Claude Opus 5 and GPT-5.2 across agentic, coding, multimodal, knowledge, reasoning, multilingual, and math tasks.

- Canonical page: https://benchlm.ai/compare/claude-opus-5-vs-gpt-5-2
- Last updated: September 4, 2026

- Shared sourced benchmarks: 4
- HTML indexing: indexable
- Ranking lane: BenchAlign v5

## Quick Verdict

Pick Claude Opus 5 if you want the stronger benchmark profile. GPT-5.2 only makes more sense when its price, context window, or workload-specific category wins matter more than the overall score.

## Summary

- Claude Opus 5 leads overall 80.66 to 65.45.
- The clearest category separation is in agentic, where the averages are 77.4 for Claude Opus 5 and 43 for GPT-5.2.
- The biggest single benchmark swing is ARC-AGI-2 in Reasoning, with scores of 90.4% and 52.9%.
- GPT-5.2 is the cheaper option on output tokens, which matters if you expect large responses or heavy interactive use.
- GPT-5.2 also has the larger context window at 400K.

## Model Snapshot

| Property | Claude Opus 5 | GPT-5.2 |
|----------|----------|----------|
| Creator | Anthropic | OpenAI |
| Type | Proprietary | Proprietary |
| Reasoning | Reasoning | Reasoning |
| Context | null | 400K |
| Overall Score | 80.66 | 65.45 |
| Benchmarks Covered | 73 | 15 |
| Pricing (input/output) | $5.00 / $25.00 | $1.75 / $14.00 |

## Category Breakdown

### Agentic

- Winner: Claude Opus 5
- Claude Opus 5 public-lane score: 77.4 (Supported · #2/151)
- GPT-5.2 public-lane score: 43 (Supported · #108/151)

| Benchmark | Claude Opus 5 | GPT-5.2 | Winner |
|-----------|-----------|-----------|--------|
| Terminal-Bench 3.0 | 42.7% | Coming soon | Coming soon |
| BrowseComp | 90.8% | 65.8% | Claude Opus 5 |
| HLE w/ tools | 64.7% | Coming soon | Coming soon |
| DeepSearchQA | 95.0% | Coming soon | Coming soon |
| DRACO | 88.6% | Coming soon | Coming soon |
| BrowseComp (10-agent, prerelease) | 93.6% | Coming soon | Coming soon |
| OSWorld 2.0 | 70.6% | Coming soon | Coming soon |
| MCP Atlas | 85.8% | Coming soon | Coming soon |
| MCP-Atlas claim coverage | 89.1% | Coming soon | Coming soon |
| LAB all-pass (Anthropic harness) | 23.58% | Coming soon | Coming soon |
| LAB criterion-pass (Anthropic harness) | 93.74% | Coming soon | Coming soon |
| LAB all-pass (Harvey held-out) | 11.7% | Coming soon | Coming soon |
| LAB criterion-pass (Harvey held-out) | 94.1% | Coming soon | Coming soon |
| Toolathlon-Verified | 80.6% | Coming soon | Coming soon |
| Toolathlon Verified Pass@3 | 87.0% | Coming soon | Coming soon |
| Toolathlon Verified Pass³ | 73.1% | Coming soon | Coming soon |
| Toolathlon Verified avg. turns | 23.5 turns | Coming soon | Coming soon |
| AutomationBench | 26.0% | Coming soon | Coming soon |
| Terminal-Bench 2.1 (Vals) | 84.6% | Coming soon | Coming soon |
| OSWorld-Verified | Coming soon | 47.3% | Coming soon |
| Gert Labs | Coming soon | 46.54% | Coming soon |
| JobBench | Coming soon | 34.3% | Coming soon |

### Coding

- Winner: Claude Opus 5
- Claude Opus 5 public-lane score: 75.6 (Supported · #3/183)
- GPT-5.2 public-lane score: 47.3 (Supported · #91/183)

| Benchmark | Claude Opus 5 | GPT-5.2 | Winner |
|-----------|-----------|-----------|--------|
| Bug Hunt Bench | 27 fixes | Coming soon | Coming soon |
| SWE-bench Verified | 96% | 80% | Claude Opus 5 |
| SWE-bench Pro | 79.2% | 55.6% | Claude Opus 5 |
| SWE Multilingual | 89.5% | Coming soon | Coming soon |
| SWE Multimodal | 59.4% | Coming soon | Coming soon |
| deepSwe | 68.8% | Coming soon | Coming soon |
| FrontierCode 1.1 Main | 53.4% | Coming soon | Coming soon |
| FrontierCode 1.1 Extended | 63.6% | Coming soon | Coming soon |
| FrontierSWE v2 | 52.0% | Coming soon | Coming soon |
| ProgramBench (episode 1) | 83.0% | Coming soon | Coming soon |
| ProgramBench | 93.0% | Coming soon | Coming soon |
| cursorBench32 | 70.0% | Coming soon | Coming soon |
| VulcanBench v3 | 87.0% | Coming soon | Coming soon |
| VulcanBench CII v1 | 96.4% | Coming soon | Coming soon |
| LiveCodeBench (Vals) | 89.0% | Coming soon | Coming soon |
| SWE-bench (Vals) | 97.0% | Coming soon | Coming soon |
| Vibe Code Bench | Coming soon | 53.50% | Coming soon |

### Multimodal & Grounded

- Winner: Coming soon
- Claude Opus 5 public-lane score: 88.7 (#2/48)
- GPT-5.2 public-lane score: 66.3 (#23/48)

| Benchmark | Claude Opus 5 | GPT-5.2 | Winner |
|-----------|-----------|-----------|--------|
| Chartography (no tools) | 29.6% | Coming soon | Coming soon |
| Chartography (tools) | 83.0% | Coming soon | Coming soon |
| BenchCAD Vision2Code (no tools) | 0.366 | Coming soon | Coming soon |
| BenchCAD Vision2Code (tools) | 0.821 | Coming soon | Coming soon |
| GDP.pdf (no tools) | 83.4% | Coming soon | Coming soon |
| GDP.pdf (tools) | 85.5% | Coming soon | Coming soon |
| OfficeQA | 78.1% | Coming soon | Coming soon |
| OfficeQA Pro | 66.9% | Coming soon | Coming soon |
| MMMU-Pro | Coming soon | 79.5% | Coming soon |
| MathVision | Coming soon | 83.0% | Coming soon |
| CharXiv | Coming soon | 82.1% | Coming soon |
| V* | Coming soon | 75.9% | Coming soon |

### Reasoning

- Winner: Coming soon
- Claude Opus 5 public-lane score: 75.6 (#13/22)
- GPT-5.2 public-lane score: 53.8 (Unranked · 3 rankable rows)

| Benchmark | Claude Opus 5 | GPT-5.2 | Winner |
|-----------|-----------|-----------|--------|
| ARC-AGI-1 | 97.50% | Coming soon | Coming soon |
| ARC-AGI-2 | 90.4% | 52.9% | Claude Opus 5 |
| ARC-AGI-3 | 30.2% | Coming soon | Coming soon |

### Knowledge

- Winner: Claude Opus 5
- Claude Opus 5 public-lane score: 82.1 (Supported · #3/181)
- GPT-5.2 public-lane score: 62 (Supported · #30/181)

| Benchmark | Claude Opus 5 | GPT-5.2 | Winner |
|-----------|-----------|-----------|--------|
| HLE | 64.7% | Coming soon | Coming soon |
| HLE w/o tools | 56.3% | Coming soon | Coming soon |
| HLE-Verified | 54.4% | Coming soon | Coming soon |
| LABBench2 | 84.2% | Coming soon | Coming soon |
| HealthBench (raw) | 67.1% | Coming soon | Coming soon |
| HealthBench (length-adjusted) | 57.8% | Coming soon | Coming soon |
| HealthBench Professional | 59.8% | Coming soon | Coming soon |
| HealthBench Professional (raw) | 73.4% | Coming soon | Coming soon |
| BioMysteryBench (human-solvable) | 90.1% | Coming soon | Coming soon |
| BioMysteryBench (human-difficult) | 49.4% | Coming soon | Coming soon |
| SpatialBench Verified | 72.5% | Coming soon | Coming soon |
| SingleCellBench | 60.6% | Coming soon | Coming soon |
| ProteinGym Hard | 47.7% | Coming soon | Coming soon |
| Protein Design | 42.5% | Coming soon | Coming soon |
| Organic chemistry V2 | 61.6% | Coming soon | Coming soon |
| Protocols (troubleshooting) | 61.1% | Coming soon | Coming soon |
| Protocols (understanding) | 78.4% | Coming soon | Coming soon |
| GPQA Diamond (Vals) | 93.4% | Coming soon | Coming soon |
| MMLU-Pro (Vals) | 91.6% | Coming soon | Coming soon |
| GPQA | Coming soon | 92.4% | Coming soon |

### Instruction Following

- Winner: Coming soon
- Claude Opus 5 public-lane score: Coming soon
- GPT-5.2 public-lane score: 92.3 (#14/120)

### Multilingual

- Winner: Coming soon
- Claude Opus 5 public-lane score: Coming soon
- GPT-5.2 public-lane score: Coming soon

| Benchmark | Claude Opus 5 | GPT-5.2 | Winner |
|-----------|-----------|-----------|--------|
| GMMLU | 92.5% | Coming soon | Coming soon |
| MILU | 92.1% | Coming soon | Coming soon |
| INCLUDE | 89.8% | Coming soon | Coming soon |

### Mathematics

- Winner: Coming soon
- Claude Opus 5 public-lane score: Coming soon
- GPT-5.2 public-lane score: 57.5 (Unranked · 2 rankable rows)

| Benchmark | Claude Opus 5 | GPT-5.2 | Winner |
|-----------|-----------|-----------|--------|
| IMO 2026 | 42/42 | Coming soon | Coming soon |
| RiemannBench (no tools) | 60.0% | Coming soon | Coming soon |
| RiemannBench (tools) | 79.0% | Coming soon | Coming soon |
| ArXivMath Jun. 2026 (no tools) | 90.8% | Coming soon | Coming soon |
| ArXivMath Jun. 2026 (tools) | 91.3% | Coming soon | Coming soon |
| FrontierMath v2 (Tiers 1-3) | Coming soon | 40.700% | Coming soon |
| FrontierMath v2 (Tier 4) | Coming soon | 18.800% | Coming soon |

## FAQ

### Which is better overall, Claude Opus 5 or GPT-5.2?

Claude Opus 5 is ahead overall on BenchLM right now.

### Where is the biggest gap between Claude Opus 5 and GPT-5.2?

The widest category gap is in agentic, where the averages are 77.4 for Claude Opus 5 and 43 for GPT-5.2.

### How many benchmarks does Claude Opus 5 cover on BenchLM?

Claude Opus 5 currently has 73 sourced benchmark scores on BenchLM.

### How many benchmarks does GPT-5.2 cover on BenchLM?

GPT-5.2 currently has 15 sourced benchmark scores on BenchLM.

## Related Comparisons

- [Claude Opus 5 vs GPT-5.2 Pro](/compare/claude-opus-5-vs-gpt-5-2-pro)
- [GPT-5.2 vs GPT-5.2 Pro](/compare/gpt-5-2-vs-gpt-5-2-pro)
- [Claude Opus 5 vs GPT-5.2 Instant](/compare/claude-opus-5-vs-gpt-5-2-instant)
- [GPT-5.2 vs GPT-5.2 Instant](/compare/gpt-5-2-vs-gpt-5-2-instant)

## Explore More

- [Claude Opus 5 profile](/models/claude-opus-5)
- [GPT-5.2 profile](/models/gpt-5-2)
- [Compare Pricing](/llm-pricing)
- [Alternative Finder](/tools/alternative-finder)
- [LLM Selector](/tools/llm-selector)
- [Overall Rankings](/best/overall)