# DeepSeek-V3.2-Exp vs GPT-4.1

> DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.9 to 35.7 on the Noometry Index.

- Canonical page: https://noometry.com/compare/deepseek-v3-2-exp-vs-gpt-4-1
- Last updated: 2026-10-10
- Shared benchmarks: 35

## Summary

- They share 35 benchmarks with published results for both. DeepSeek-V3.2-Exp scores higher in 9 categories and GPT-4.1 in 0 categories; 9 gaps are clear of the uncertainty.
- The widest gap is in math, where DeepSeek-V3.2-Exp leads 40.8 to 21.4.
- The biggest single-benchmark swing is ARC-AGI-1: 57% for DeepSeek-V3.2-Exp and 5.5% for GPT-4.1.
- DeepSeek-V3.2-Exp is cheaper at $0.26 / $0.38 per million input/output tokens, against $2 / $8 for GPT-4.1.
- GPT-4.1 accepts more context: 1.05M tokens versus 164K.
- DeepSeek-V3.2-Exp has downloadable open weights; the other is API-only.

## Snapshot

| | DeepSeek-V3.2-Exp | GPT-4.1 |
|---|---|---|
| Provider | DeepSeek | OpenAI |
| Noometry Index | 44.9 | 35.7 |
| Rank | 68 | 215 |
| Context | 164K | 1.05M |
| Input $/M | $0.26 | $2 |
| Output $/M | $0.38 | $8 |
| Weights | Open | Proprietary |

## Coding

- DeepSeek-V3.2-Exp: 46.4 (#61)
- GPT-4.1: 34.2 (#228)

| Benchmark | DeepSeek-V3.2-Exp | GPT-4.1 |
|---|---|---|
| SWE-bench Verified (bash only) | 70% | 39.6% |
| Aider Polyglot | 74.2% | 52.4% |
| WeirdML | 39.5% | 39% |
| LMArena Coding | 1454 | 1391 |
| SWE-bench Verified | — | 48.5% |
| LMArena WebDev | 1362 | — |
| SWE-bench Multilingual | 59% | — |
| SciCode | 38.9% | — |
| CadEval | — | 42% |
| ALE-Bench | — | 558.1 |

## Agentic & Tool Use

- DeepSeek-V3.2-Exp: 37.9 (#31)
- GPT-4.1: 34.4 (#40)

| Benchmark | DeepSeek-V3.2-Exp | GPT-4.1 |
|---|---|---|
| Berkeley Function Calling Leaderboard | 56.7% | 54% |
| Terminal-Bench | 39.6% | — |
| TheAgentCompany | 42.9% | — |
| Vending-Bench 2 | 1,034 | — |

## Reasoning

- DeepSeek-V3.2-Exp: 22.0 (#185)
- GPT-4.1: 11.7 (#334)

| Benchmark | DeepSeek-V3.2-Exp | GPT-4.1 |
|---|---|---|
| ARC-AGI-2 | 4% | 0.4% |
| Kagi LLM Benchmark | 52.2% | 52.3% |
| ARC-AGI-1 | 57% | 5.5% |
| Chess Puzzles | 14% | 6% |
| LMArena Hard Prompts | 1433 | 1384 |
| DTBench | 87.7% | 68.3% |
| LMCA | 29.1% | 25.6% |
| Epoch Capabilities Index | 146.28 | 136.79 |
| SimpleBench | — | 27% |
| NYT Connections (extended) | 36.7% | — |
| CritPt | 2.9% | — |
| EnigmaEval | — | 2.2% |
| Thematic Generalization | 65% | — |
| ForecastBench | — | 61.5 |

## Math

- DeepSeek-V3.2-Exp: 40.8 (#75)
- GPT-4.1: 21.4 (#269)

| Benchmark | DeepSeek-V3.2-Exp | GPT-4.1 |
|---|---|---|
| OTIS Mock AIME 2024-2025 | 87.8% | 38.3% |
| LMArena Math | 1435 | 1369 |
| FrontierMath (Feb 2025 set) | 22.1% | 5.5% |
| FrontierMath Tier 4 (v1) | 2.1% | 0% |
| FrontierMath (Tiers 1-3) | — | 6% |
| MathArena Final-Answer Competitions | 57.7% | — |
| ProofBench | 8% | — |
| Omni-MATH | — | 47.1% |
| MATH Level 5 | — | 83% |

## Knowledge

- DeepSeek-V3.2-Exp: 51.6 (#61)
- GPT-4.1: 37.0 (#152)

| Benchmark | DeepSeek-V3.2-Exp | GPT-4.1 |
|---|---|---|
| GPQA Diamond | 83.4% | 66.9% |
| Vectara Hallucination Rate | 5.3% | 5.6% |
| LMArena Expert | 1436 | 1364 |
| Humanity's Last Exam | — | 5.4% |
| SimpleQA Verified | — | 31.1% |
| MMLU-Pro | — | 81.1% |
| GPQA (HELM) | — | 65.9% |

## Multimodal

- DeepSeek-V3.2-Exp: —
- GPT-4.1: 37.5 (#66)

| Benchmark | DeepSeek-V3.2-Exp | GPT-4.1 |
|---|---|---|
| LMArena Vision | — | 1210 |
| GeoBench | — | 72% |

## Multilingual

- DeepSeek-V3.2-Exp: 52.4 (#85)
- GPT-4.1: 49.6 (#128)

| Benchmark | DeepSeek-V3.2-Exp | GPT-4.1 |
|---|---|---|
| LMArena Non-English | 1409 | 1370 |
| LMArena Chinese | 1461 | 1382 |
| LMArena French | 1435 | 1383 |
| LMArena German | 1440 | 1380 |
| LMArena Japanese | 1377 | 1320 |
| LMArena Korean | 1372 | 1339 |
| LMArena Russian | 1423 | 1377 |
| LMArena Spanish | 1441 | 1375 |

## Instruction Following

- DeepSeek-V3.2-Exp: 74.7 (#89)
- GPT-4.1: 71.5 (#148)

| Benchmark | DeepSeek-V3.2-Exp | GPT-4.1 |
|---|---|---|
| LMArena Instruction Following | 1413 | 1367 |
| IFEval | — | 83.8% |

## Long Context

- DeepSeek-V3.2-Exp: 47.6 (#15)
- GPT-4.1: 40.0 (#158)

| Benchmark | DeepSeek-V3.2-Exp | GPT-4.1 |
|---|---|---|
| Fiction.LiveBench | 83.3% | 63.9% |
| LMArena Longer Query | 1428 | 1385 |
| CL-bench | 13.2% | — |
| CL-bench Life | 9.5% | — |

## Writing & Preference

- DeepSeek-V3.2-Exp: 62.5 (#73)
- GPT-4.1: 57.8 (#120)

| Benchmark | DeepSeek-V3.2-Exp | GPT-4.1 |
|---|---|---|
| LMArena Text | 1425 | 1383 |
| LMArena Creative Writing | 1403 | 1363 |
| EQ-Bench Creative Writing | 1515 | 1420 |
| LMArena Multi-Turn | 1427 | 1398 |
| WildBench | — | 85.4% |

## FAQ

### Is DeepSeek-V3.2-Exp better than GPT-4.1?

DeepSeek-V3.2-Exp is the stronger model overall, scoring 44.9 to 35.7 on the Noometry Index.

### Which is cheaper, DeepSeek-V3.2-Exp or GPT-4.1?

DeepSeek-V3.2-Exp is cheaper. It lists at $0.26 per million input tokens and $0.38 per million output tokens; GPT-4.1 lists at $2 and $8.

### Is DeepSeek-V3.2-Exp or GPT-4.1 better for coding?

DeepSeek-V3.2-Exp scores higher on coding benchmarks: 46.4 versus 34.2 in the Noometry coding category.

### Which has the bigger context window?

GPT-4.1 does, with 1.05M tokens against 164K.

### How many benchmarks do DeepSeek-V3.2-Exp and GPT-4.1 share?

35 benchmarks have published results for both models. DeepSeek-V3.2-Exp has 48 scored results on Noometry and GPT-4.1 has 52.