# Claude Sonnet 4.6 vs GPT-5.4 nano

> Side-by-side benchmark comparison for Claude Sonnet 4.6 and GPT-5.4 nano across agentic, coding, multimodal, knowledge, reasoning, multilingual, and math tasks.

- Canonical page: https://benchlm.ai/compare/claude-sonnet-4-6-vs-gpt-5-4-nano
- Last updated: September 10, 2026

- Shared sourced benchmarks: 12
- HTML indexing: indexable
- Ranking lane: BenchAlign v5

## Quick Verdict

Pick Claude Sonnet 4.6 if you want the stronger benchmark profile. GPT-5.4 nano only makes more sense when its price, context window, or workload-specific category wins matter more than the overall score.

## Summary

- Claude Sonnet 4.6 leads overall 62.96 to 59.57.
- The clearest category separation is in instruction following, where the averages are 48.2 for Claude Sonnet 4.6 and 93.2 for GPT-5.4 nano.
- The biggest single benchmark swing is OSWorld-Verified in Agentic, with scores of 72.1% and 39%.
- GPT-5.4 nano is the cheaper option on output tokens, which matters if you expect large responses or heavy interactive use.
- GPT-5.4 nano also has the larger context window at 400K.

## Model Snapshot

| Property | Claude Sonnet 4.6 | GPT-5.4 nano |
|----------|----------|----------|
| Creator | Anthropic | OpenAI |
| Type | Proprietary | Proprietary |
| Reasoning | Non-Reasoning | Reasoning |
| Context | 200K | 400K |
| Overall Score | 62.96 | 59.57 |
| Benchmarks Covered | 26 | 18 |
| Pricing (input/output) | $3.00 / $15.00 | $0.20 / $1.25 |

## Category Breakdown

### Agentic

- Winner: Claude Sonnet 4.6
- Claude Sonnet 4.6 public-lane score: 44 (Supported · #93/152)
- GPT-5.4 nano public-lane score: 34.6 (Supported · #133/152)

| Benchmark | Claude Sonnet 4.6 | GPT-5.4 nano | Winner |
|-----------|-----------|-----------|--------|
| Terminal-Bench 2.0 | 59.1% | 46.3% | Claude Sonnet 4.6 |
| OSWorld-Verified | 72.1% | 39% | Claude Sonnet 4.6 |
| Claw-Eval | 67.8% | Coming soon | Coming soon |
| CyberGym | 65.2% | Coming soon | Coming soon |
| Gert Labs | 62.92% | Coming soon | Coming soon |
| OSWorld 2.0 | 8.3% | Coming soon | Coming soon |
| JobBench | 36.9% | Coming soon | Coming soon |
| Terminal-Bench 2.1 (Vals) | 57.3% | 41.6% | Claude Sonnet 4.6 |
| ApprenticeBench | 2% | Coming soon | Coming soon |
| MCP Atlas | Coming soon | 56.1% | Coming soon |
| Toolathlon | Coming soon | 35.5% | Coming soon |
| τ²-bench results | Coming soon | 92.5% | Coming soon |

### Coding

- Winner: Claude Sonnet 4.6
- Claude Sonnet 4.6 public-lane score: 52.2 (Supported · #48/151)
- GPT-5.4 nano public-lane score: 37.1 (Supported · #126/151)

| Benchmark | Claude Sonnet 4.6 | GPT-5.4 nano | Winner |
|-----------|-----------|-----------|--------|
| SWE-bench Verified | 79.6% | Coming soon | Coming soon |
| SWE-Rebench | 60.7% | Coming soon | Coming soon |
| React Native Evals | 80.6% | Coming soon | Coming soon |
| Vibe Code Bench | 51.48% | 26.10% | Claude Sonnet 4.6 |
| cursorBench31 | 48.8% | Coming soon | Coming soon |
| FrontierCode 1.1 Main | 24.3% | Coming soon | Coming soon |
| LiveCodeBench (Vals) | 82.1% | 84.0% | GPT-5.4 nano |
| SWE-bench (Vals) | 77.4% | 69.8% | Claude Sonnet 4.6 |

### Multimodal & Grounded

- Winner: Coming soon
- Claude Sonnet 4.6 public-lane score: 54.1 (#33/48)
- GPT-5.4 nano public-lane score: 23.8 (#45/48)

| Benchmark | Claude Sonnet 4.6 | GPT-5.4 nano | Winner |
|-----------|-----------|-----------|--------|
| CharXiv | 77.4% | Coming soon | Coming soon |
| MMMU-Pro | Coming soon | 66.1% | Coming soon |
| MMMU-Pro w/ Python | Coming soon | 69.5% | Coming soon |

### Reasoning

- Winner: Coming soon
- Claude Sonnet 4.6 public-lane score: 67.9 (Unranked · 2 rankable rows)
- GPT-5.4 nano public-lane score: 73.7 (Unranked · 2 rankable rows)

### Knowledge

- Winner: Claude Sonnet 4.6
- Claude Sonnet 4.6 public-lane score: 55.8 (Supported · #51/183)
- GPT-5.4 nano public-lane score: 47.4 (Supported · #97/183)

| Benchmark | Claude Sonnet 4.6 | GPT-5.4 nano | Winner |
|-----------|-----------|-----------|--------|
| GPQA | 89.9% | 82.8% | Claude Sonnet 4.6 |
| SuperGPQA | 95% | Coming soon | Coming soon |
| MMLU-Pro | 79.2% | Coming soon | Coming soon |
| HLE | 49% | 37.7% | Claude Sonnet 4.6 |
| GPQA Diamond (Vals) | 85.6% | 77.5% | Claude Sonnet 4.6 |
| MMLU-Pro (Vals) | 87.3% | 77.2% | Claude Sonnet 4.6 |
| HLE w/o tools | Coming soon | 24.3% | Coming soon |

### Instruction Following

- Winner: Coming soon
- Claude Sonnet 4.6 public-lane score: 48.2 (#86/123)
- GPT-5.4 nano public-lane score: 93.2 (#9/123)

### Mathematics

- Winner: Coming soon
- Claude Sonnet 4.6 public-lane score: 49 (Unranked · 2 rankable rows)
- GPT-5.4 nano public-lane score: 43.9 (Unranked · 2 rankable rows)

| Benchmark | Claude Sonnet 4.6 | GPT-5.4 nano | Winner |
|-----------|-----------|-----------|--------|
| FrontierMath v2 (Tiers 1-3) | 32.400% | 25.860% | Claude Sonnet 4.6 |
| FrontierMath v2 (Tier 4) | 8.300% | 6.250% | Claude Sonnet 4.6 |

## FAQ

### Which is better overall, Claude Sonnet 4.6 or GPT-5.4 nano?

Claude Sonnet 4.6 is ahead overall on BenchLM right now.

### Where is the biggest gap between Claude Sonnet 4.6 and GPT-5.4 nano?

The widest category gap is in instruction following, where the averages are 48.2 for Claude Sonnet 4.6 and 93.2 for GPT-5.4 nano.

### How many benchmarks does Claude Sonnet 4.6 cover on BenchLM?

Claude Sonnet 4.6 currently has 26 sourced benchmark scores on BenchLM.

### How many benchmarks does GPT-5.4 nano cover on BenchLM?

GPT-5.4 nano currently has 18 sourced benchmark scores on BenchLM.

## Related Comparisons

- [Claude Sonnet 4.6 vs GPT-5.4](/compare/claude-sonnet-4-6-vs-gpt-5-4)
- [GPT-5.4 nano vs GPT-5.4](/compare/gpt-5-4-vs-gpt-5-4-nano)
- [Claude Sonnet 4.6 vs GPT-5.4 mini](/compare/claude-sonnet-4-6-vs-gpt-5-4-mini)
- [GPT-5.4 nano vs GPT-5.4 mini](/compare/gpt-5-4-mini-vs-gpt-5-4-nano)

## Explore More

- [Claude Sonnet 4.6 profile](/models/claude-sonnet-4-6)
- [GPT-5.4 nano profile](/models/gpt-5-4-nano)
- [Compare Pricing](/llm-pricing)
- [Alternative Finder](/tools/alternative-finder)
- [LLM Selector](/tools/llm-selector)
- [Overall Rankings](/best/overall)