# Claude Sonnet 4.6 vs GPT-5.2-Codex

> Side-by-side benchmark comparison for Claude Sonnet 4.6 and GPT-5.2-Codex across agentic, coding, multimodal, knowledge, reasoning, multilingual, and math tasks.

- Canonical page: https://benchlm.ai/compare/claude-sonnet-4-6-vs-gpt-5-2-codex
- Last updated: October 17, 2026

- Shared sourced benchmarks: 5
- HTML indexing: indexable
- Ranking lane: BenchAlign v5.8

## Quick Verdict

Claude Sonnet 4.6 leads by point estimate. Conditional score ranges do not establish rank confidence. Choose using the category evidence and your own workload tests.

## Summary

- Claude Sonnet 4.6 has the higher overall point estimate 55.89 to 55.3. Conditional score ranges do not establish rank confidence.
- The clearest category separation is in instruction following, where the averages are 48 for Claude Sonnet 4.6 and 93.8 for GPT-5.2-Codex.
- The biggest single benchmark swing is Vibe Code Bench in Coding, with scores of 51.48% and 37.91%.
- GPT-5.2-Codex is the cheaper option on output tokens, which matters if you expect large responses or heavy interactive use.
- GPT-5.2-Codex also has the larger context window at 400K.

## Model Snapshot

| Property | Claude Sonnet 4.6 | GPT-5.2-Codex |
|----------|----------|----------|
| Creator | Anthropic | OpenAI |
| Type | Proprietary | Proprietary |
| Reasoning | Non-Reasoning | Reasoning |
| Context | 200K | 400K |
| Overall Score | 55.89 | 55.3 |
| Benchmarks Covered | 28 | 5 |
| Pricing (input/output) | $3.00 / $15.00 | $1.75 / $14.00 |

## Category Breakdown

### Agentic

- Winner: Directional only
- Claude Sonnet 4.6 public-lane score: 46.3 (Supported · #55/128)
- GPT-5.2-Codex public-lane score: 40.8 (Estimated · #64/128)

| Benchmark | Claude Sonnet 4.6 | GPT-5.2-Codex | Winner |
|-----------|-----------|-----------|--------|
| Terminal-Bench 2.0 | 59.1% | Coming soon | Coming soon |
| OSWorld-Verified | 72.1% | Coming soon | Coming soon |
| Claw-Eval | 67.8% | Coming soon | Coming soon |
| CyberGym | 65.2% | Coming soon | Coming soon |
| Gert Labs | 62.92% | 51.79% | Claude Sonnet 4.6 |
| OSWorld 2.0 | 8.3% | Coming soon | Coming soon |
| JobBench | 36.9% | 26.0% | Claude Sonnet 4.6 |
| Terminal-Bench 2.1 (Vals) | 57.3% | Coming soon | Coming soon |
| ApprenticeBench | 2% | Coming soon | Coming soon |
| MazeBench | 0% | Coming soon | Coming soon |

### Coding

- Winner: Claude Sonnet 4.6
- Claude Sonnet 4.6 public-lane score: 44.7 (Supported · #54/150)
- GPT-5.2-Codex public-lane score: 43.7 (Supported · #56/150)

| Benchmark | Claude Sonnet 4.6 | GPT-5.2-Codex | 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 |
| Terminal-Bench 2.0 | 59.1% | Coming soon | Coming soon |
| Vibe Code Bench | 51.48% | 37.91% | 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% | 88.0% | GPT-5.2-Codex |
| SWE-bench (Vals) | 77.4% | 72.4% | Claude Sonnet 4.6 |

### Reasoning

- Winner: Not comparable
- Claude Sonnet 4.6 public-lane score: 62.8 (Unranked · 4 rankable rows)
- GPT-5.2-Codex public-lane score: 82.4 (Unranked · 2 rankable rows)

| Benchmark | Claude Sonnet 4.6 | GPT-5.2-Codex | Winner |
|-----------|-----------|-----------|--------|
| ARC-AGI-2 | 58.3% | Coming soon | Coming soon |
| ARC-AGI-1 | 86.00% | Coming soon | Coming soon |

### Multimodal & Grounded

- Winner: Not comparable
- Claude Sonnet 4.6 public-lane score: 58.7 (#45/54)
- GPT-5.2-Codex public-lane score: 76.9 (Unranked · 1 rankable row)

| Benchmark | Claude Sonnet 4.6 | GPT-5.2-Codex | Winner |
|-----------|-----------|-----------|--------|
| CharXiv | 77.4% | Coming soon | Coming soon |

### Knowledge

- Winner: Directional only
- Claude Sonnet 4.6 public-lane score: 55.4 (Supported · #58/184)
- GPT-5.2-Codex public-lane score: 56.4 (Estimated · #54/184)

| Benchmark | Claude Sonnet 4.6 | GPT-5.2-Codex | Winner |
|-----------|-----------|-----------|--------|
| GPQA | 89.9% | Coming soon | Coming soon |
| SuperGPQA | 95% | Coming soon | Coming soon |
| MMLU-Pro | 79.2% | Coming soon | Coming soon |
| HLE | 49% | Coming soon | Coming soon |
| GPQA Diamond (Vals) | 85.6% | Coming soon | Coming soon |
| MMLU-Pro (Vals) | 87.3% | Coming soon | Coming soon |

### Instruction Following

- Winner: Directional only
- Claude Sonnet 4.6 public-lane score: 48 (#99/148)
- GPT-5.2-Codex public-lane score: 93.8 (#4/148)

## FAQ

### Which is better overall, Claude Sonnet 4.6 or GPT-5.2-Codex?

Claude Sonnet 4.6 has the higher overall point estimate. Conditional score ranges do not establish rank confidence.

### Where is the biggest gap between Claude Sonnet 4.6 and GPT-5.2-Codex?

The widest category gap is in instruction following, where the averages are 48 for Claude Sonnet 4.6 and 93.8 for GPT-5.2-Codex.

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

Claude Sonnet 4.6 currently has 28 sourced benchmark scores on BenchLM.

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

GPT-5.2-Codex currently has 5 sourced benchmark scores on BenchLM.

## Related Comparisons

- [Claude Sonnet 4.6 vs Claude Opus 5.5](/compare/claude-opus-5-5-vs-claude-sonnet-4-6)
- [GPT-5.2-Codex vs Claude Opus 5.5](/compare/claude-opus-5-5-vs-gpt-5-2-codex)
- [Claude Sonnet 4.6 vs GPT-6.1 Sol](/compare/claude-sonnet-4-6-vs-gpt-6-1-sol)
- [GPT-5.2-Codex vs GPT-6.1 Sol](/compare/gpt-5-2-codex-vs-gpt-6-1-sol)

## Explore More

- [Claude Sonnet 4.6 profile](/models/claude-sonnet-4-6)
- [GPT-5.2-Codex profile](/models/gpt-5-2-codex)
- [Compare Pricing](/llm-pricing)
- [Alternative Finder](/tools/alternative-finder)
- [LLM Selector](/tools/llm-selector)
- [Overall Rankings](/best/overall)