# DeepSeek V3 vs DeepSeek V3.2

> Side-by-side benchmark comparison for DeepSeek V3 and DeepSeek V3.2 across agentic, coding, multimodal, knowledge, reasoning, multilingual, and math tasks.

- Canonical page: https://benchlm.ai/compare/deepseek-v3-vs-deepseek-v3-2
- Last updated: September 29, 2026

- Shared sourced benchmarks: 1
- HTML indexing: indexable
- Ranking lane: BenchAlign v5.7

## Quick Verdict

Evidence is too thin for a benchmark verdict: DeepSeek V3 and DeepSeek V3.2 share only 1 sourced benchmark row.

## Summary

- Only 1 shared sourced benchmark row is available. That is not enough for a broad capability verdict, so this mirror does not name an overall winner.
- DeepSeek V3.2 has the lower published output-token price.

## Model Snapshot

| Property | DeepSeek V3 | DeepSeek V3.2 |
|----------|----------|----------|
| Creator | DeepSeek | DeepSeek |
| Type | Open Weight | Open Weight |
| Reasoning | Non-Reasoning | Non-Reasoning |
| Context | 128K | 128K |
| Independent public score (not head-to-head) | 31.86 | 50.13 |
| Benchmarks Covered | 6 | 7 |
| Pricing (input/output) | $0.27 / $1.10 | $0.28 / $0.42 |

## Category Breakdown

### Agentic

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: 12.1 (Estimated · #113/117)
- DeepSeek V3.2 public-lane score: Coming soon

| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Winner |
|-----------|-----------|-----------|--------|
| Claw-Eval | Coming soon | 40.2% | Coming soon |
| VITA-Bench | Coming soon | 18.5% | Coming soon |
| Gert Labs | Coming soon | 29.57% | Coming soon |

### Coding

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: 18.6 (Estimated · #131/143)
- DeepSeek V3.2 public-lane score: 32.1 (Estimated · #89/143)

| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Winner |
|-----------|-----------|-----------|--------|
| LiveCodeBench | 37.6% | Coming soon | Coming soon |
| SWE-bench Verified | 42% | Coming soon | Coming soon |
| SWE-Rebench | Coming soon | 60.9% | Coming soon |
| React Native Evals | Coming soon | 71.5% | Coming soon |

### Reasoning

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: 42.3 (Unranked · 2 rankable rows)
- DeepSeek V3.2 public-lane score: 53.6 (Unranked · 2 rankable rows)

### Knowledge

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: 30.2 (Estimated · #140/169)
- DeepSeek V3.2 public-lane score: 41.7 (Estimated · #91/169)

| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Winner |
|-----------|-----------|-----------|--------|
| GPQA | 59.1% | Coming soon | Coming soon |
| MMLU-Pro | 75.9% | Coming soon | Coming soon |

### Instruction Following

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: 38.2 (#103/124)
- DeepSeek V3.2 public-lane score: 56.7 (#75/124)

| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Winner |
|-----------|-----------|-----------|--------|
| IFEval | 86.1% | Coming soon | Coming soon |

### Mathematics

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: 26 (Unranked · 1 rankable row)
- DeepSeek V3.2 public-lane score: 40 (Unranked · 2 rankable rows)

| Benchmark | DeepSeek V3 | DeepSeek V3.2 | Winner |
|-----------|-----------|-----------|--------|
| FrontierMath v2 (Tiers 1-3) | 1.724% | 22.100% | DeepSeek V3.2 |
| FrontierMath v2 (Tier 4) | Coming soon | 2.100% | Coming soon |

## FAQ

### Is there enough evidence to pick between DeepSeek V3 and DeepSeek V3.2?

Not yet. The models share only 1 sourced benchmark row, so this mirror reports the available measurements and metadata without naming a broad winner.

### What can I compare safely on this page?

Published pricing, context windows, model type, and individual shared benchmark rows are directly comparable. Independent overall scores and non-overlapping benchmark coverage are labeled separately and should not be read as a controlled head-to-head result.

## Related Comparisons

- [DeepSeek V3 vs DeepSeek V3.2 (Thinking)](/compare/deepseek-v3-vs-deepseek-v3-2-thinking)
- [DeepSeek V3.2 vs DeepSeek V3.2 (Thinking)](/compare/deepseek-v3-2-vs-deepseek-v3-2-thinking)
- [DeepSeek V3 vs GPT-6 Astra](/compare/deepseek-v3-vs-gpt-6-astra)
- [DeepSeek V3.2 vs GPT-6 Astra](/compare/deepseek-v3-2-vs-gpt-6-astra)

## Explore More

- [DeepSeek V3 profile](/models/deepseek-v3)
- [DeepSeek V3.2 profile](/models/deepseek-v3-2)
- [Compare Pricing](/llm-pricing)
- [Alternative Finder](/tools/alternative-finder)
- [LLM Selector](/tools/llm-selector)
- [Overall Rankings](/best/overall)