# DeepSeek-V3 vs Qwen3.8 27B

> Qwen3.8 27B is the stronger model overall, scoring 45.3 to 39.2 on the Noometry Index. DeepSeek-V3 costs 2.8× less per token, which makes it the better buy when Qwen3.8 27B's lead doesn't matter for your workload.

- Canonical page: https://noometry.com/compare/deepseek-v3-vs-qwen3-8-27b
- Last updated: 2026-10-10
- Shared benchmarks: 23

## Summary

- They share 23 benchmarks with published results for both. DeepSeek-V3 scores higher in 0 categories and Qwen3.8 27B in 8 categories; 8 gaps are clear of the uncertainty.
- The widest gap is in reasoning, where Qwen3.8 27B leads 31.5 to 20.0.
- The biggest single-benchmark swing is LMCA: 15.5% for DeepSeek-V3 and 41.4% for Qwen3.8 27B.
- DeepSeek-V3 is cheaper at $0.24 / $0.90 per million input/output tokens, against $0.99 / $1.49 for Qwen3.8 27B.
- Qwen3.8 27B accepts more context: 262K tokens versus 164K.

## Snapshot

| | DeepSeek-V3 | Qwen3.8 27B |
|---|---|---|
| Provider | DeepSeek | Alibaba (Qwen) |
| Noometry Index | 39.2 | 45.3 |
| Rank | 157 | 64 |
| Context | 164K | 262K |
| Input $/M | $0.24 | $0.99 |
| Output $/M | $0.90 | $1.49 |
| Weights | Open | Open |

## Coding

- DeepSeek-V3: 41.9 (#99)
- Qwen3.8 27B: 50.3 (#42)

| Benchmark | DeepSeek-V3 | Qwen3.8 27B |
|---|---|---|
| SciCode | 35.8% | 46.6% |
| LMArena Coding | 1368 | 1487 |
| Aider Polyglot | 55.1% | — |
| LMArena WebDev | — | 1592 |
| WeirdML | 36.1% | — |
| BigCodeBench Instruct | 50% | — |
| LiveBench Coding | 70.9% | — |
| BigCodeBench Complete | 62.2% | — |
| HumanEval+ | 86.6% | — |
| MBPP+ | 73% | — |

## Agentic & Tool Use

- DeepSeek-V3: —
- Qwen3.8 27B: —

| Benchmark | DeepSeek-V3 | Qwen3.8 27B |
|---|---|---|
| METR Time Horizons | 49.6% | — |

## Reasoning

- DeepSeek-V3: 20.0 (#226)
- Qwen3.8 27B: 31.5 (#71)

| Benchmark | DeepSeek-V3 | Qwen3.8 27B |
|---|---|---|
| CritPt | 0% | 5.4% |
| LMArena Hard Prompts | 1365 | 1459 |
| DTBench | 64.8% | 88% |
| LMCA | 15.5% | 41.4% |
| Epoch Capabilities Index | 135.95 | 149.38 |
| SimpleBench | 27.2% | — |
| Kagi LLM Benchmark | 52.3% | — |
| NYT Connections (extended) | — | 54.5% |
| LiveBench Reasoning | 65.8% | — |
| LiveBench Data Analysis | 60.9% | — |
| Surface Evolver Bench | — | 45% |
| BIG-Bench Hard | 87.5% | — |
| ForecastBench | 59.1 | — |
| HellaSwag | 88.9% | — |
| LiveBench | 66.9% | — |
| PIQA | 84.7% | — |
| WinoGrande | 85.2% | — |

## Math

- DeepSeek-V3: 30.4 (#208)
- Qwen3.8 27B: 35.7 (#139)

| Benchmark | DeepSeek-V3 | Qwen3.8 27B |
|---|---|---|
| LMArena Math | 1373 | 1453 |
| OTIS Mock AIME 2024-2025 | 37.8% | — |
| ProofBench | — | 16% |
| Omni-MATH | 40.3% | — |
| LiveBench Math | 73.5% | — |
| MATH Level 5 | 75.5% | — |
| FrontierMath (Feb 2025 set) | 1.7% | — |

## Knowledge

- DeepSeek-V3: 37.5 (#147)
- Qwen3.8 27B: 40.9 (#105)

| Benchmark | DeepSeek-V3 | Qwen3.8 27B |
|---|---|---|
| LMArena Expert | 1352 | 1482 |
| GPQA Diamond | 67.6% | — |
| MMLU-Pro | 72.3% | — |
| Confabulations | 26.1% | — |
| Vectara Hallucination Rate | 6.1% | — |
| GPQA (HELM) | 53.8% | — |
| ARC (AI2) Challenge | 95.3% | — |
| MMLU | 87.2% | — |
| TriviaQA | 82.9% | — |

## Multimodal

- DeepSeek-V3: —
- Qwen3.8 27B: 40.5 (#36)

| Benchmark | DeepSeek-V3 | Qwen3.8 27B |
|---|---|---|
| LMArena Vision | — | 1270 |

## Multilingual

- DeepSeek-V3: 48.8 (#138)
- Qwen3.8 27B: 54.0 (#54)

| Benchmark | DeepSeek-V3 | Qwen3.8 27B |
|---|---|---|
| LMArena Non-English | 1358 | 1431 |
| LMArena Chinese | 1390 | 1504 |
| LMArena French | 1386 | 1468 |
| LMArena German | 1374 | 1435 |
| LMArena Japanese | 1333 | 1392 |
| LMArena Korean | 1321 | 1389 |
| LMArena Russian | 1372 | 1421 |
| LMArena Spanish | 1358 | 1444 |

## Instruction Following

- DeepSeek-V3: 73.0 (#125)
- Qwen3.8 27B: 75.8 (#54)

| Benchmark | DeepSeek-V3 | Qwen3.8 27B |
|---|---|---|
| LMArena Instruction Following | 1345 | 1436 |
| LiveBench Instruction Following | 81.5% | — |
| IFEval | 83.2% | — |

## Long Context

- DeepSeek-V3: 34.0 (#248)
- Qwen3.8 27B: 44.3 (#66)

| Benchmark | DeepSeek-V3 | Qwen3.8 27B |
|---|---|---|
| LMArena Longer Query | 1353 | 1448 |
| Fiction.LiveBench | 50% | — |

## Writing & Preference

- DeepSeek-V3: 57.6 (#124)
- Qwen3.8 27B: 65.9 (#41)

| Benchmark | DeepSeek-V3 | Qwen3.8 27B |
|---|---|---|
| LMArena Text | 1375 | 1441 |
| LMArena Creative Writing | 1364 | 1382 |
| EQ-Bench Creative Writing | 1472 | 1671 |
| LMArena Multi-Turn | 1389 | 1443 |
| Short-Story Creative Writing | 77% | — |
| WildBench | 83% | — |
| LiveBench Language | 49.1% | — |

## FAQ

### Is DeepSeek-V3 better than Qwen3.8 27B?

Qwen3.8 27B is the stronger model overall, scoring 45.3 to 39.2 on the Noometry Index. DeepSeek-V3 costs 2.8× less per token, which makes it the better buy when Qwen3.8 27B's lead doesn't matter for your workload.

### Which is cheaper, DeepSeek-V3 or Qwen3.8 27B?

DeepSeek-V3 is cheaper. It lists at $0.24 per million input tokens and $0.90 per million output tokens; Qwen3.8 27B lists at $0.99 and $1.49.

### Is DeepSeek-V3 or Qwen3.8 27B better for coding?

Qwen3.8 27B scores higher on coding benchmarks: 50.3 versus 41.9 in the Noometry coding category.

### Which has the bigger context window?

Qwen3.8 27B does, with 262K tokens against 164K.

### How many benchmarks do DeepSeek-V3 and Qwen3.8 27B share?

23 benchmarks have published results for both models. DeepSeek-V3 has 60 scored results on Noometry and Qwen3.8 27B has 28.