# DeepSeek V3 vs Hy4 preview

> Side-by-side benchmark comparison for DeepSeek V3 and Hy4 preview across agentic, coding, multimodal, knowledge, reasoning, multilingual, and math tasks.

- Canonical page: https://benchlm.ai/compare/deepseek-v3-vs-hy4-preview
- Last updated: October 13, 2026

- Shared sourced benchmarks: 1
- HTML indexing: indexable
- Ranking lane: BenchAlign v5.8

## Quick Verdict

Evidence is too thin for a benchmark verdict: DeepSeek V3 and Hy4 preview 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.
- Hy4 preview has the lower published output-token price.
- Hy4 preview has the larger published context window at 1M.

## Model Snapshot

| Property | DeepSeek V3 | Hy4 preview |
|----------|----------|----------|
| Creator | DeepSeek | Tencent |
| Type | Open Weight | Open Weight |
| Reasoning | Non-Reasoning | Reasoning |
| Context | 128K | 1M |
| Independent public score (not head-to-head) | 31.77 | 60.86 |
| Benchmarks Covered | 9 | 30 |
| Pricing (input/output) | $0.27 / $1.10 | $0.00 / $0.00 |

## Category Breakdown

### Agentic

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: 10 (Estimated · #117/123)
- Hy4 preview public-lane score: 54.5 (Estimated · #37/123)

| Benchmark | DeepSeek V3 | Hy4 preview | Winner |
|-----------|-----------|-----------|--------|
| Terminal-Bench 2.1 | Coming soon | 85.4% | Coming soon |
| CyberGym | Coming soon | 78.4% | Coming soon |
| WideResearch | Coming soon | 83.9% | Coming soon |
| DRACO | Coming soon | 77.2% | Coming soon |
| MCP Atlas | Coming soon | 83.7% | Coming soon |
| Toolathlon-Verified | Coming soon | 74.1% | Coming soon |
| APEX-Agents | Coming soon | 37.1% | Coming soon |
| skillsBench | Coming soon | 62.9% | Coming soon |
| JobBench | Coming soon | 61.7% | Coming soon |
| Agents' Last Exam | Coming soon | 22.8% | Coming soon |
| AutomationBench | Coming soon | 32.1% | Coming soon |
| BankerToolBench | Coming soon | 78.6% | Coming soon |
| HLE w/ tools | Coming soon | 55.4% | Coming soon |
| Terminal-Bench 2.1 (Vals) | Coming soon | 55.1% | Coming soon |
| CWE-bench v1 | Coming soon | 53.0% | Coming soon |

### Coding

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: 15.2 (Estimated · #133/148)
- Hy4 preview public-lane score: 57.1 (Supported · #26/148)

| Benchmark | DeepSeek V3 | Hy4 preview | Winner |
|-----------|-----------|-----------|--------|
| LiveCodeBench | 37.6% | Coming soon | Coming soon |
| SWE-bench Verified | 42% | Coming soon | Coming soon |
| Terminal-Bench 2.1 | Coming soon | 85.4% | Coming soon |
| SWE-bench Pro | Coming soon | 65.7% | Coming soon |
| SWE Multilingual | Coming soon | 82.9% | Coming soon |
| DeepSWE | Coming soon | 64.3% | Coming soon |
| NL2Repo | Coming soon | 58.9% | Coming soon |
| ProgramBench | Coming soon | 17.5% | Coming soon |
| PostTrain Bench | Coming soon | 35.6% | Coming soon |
| sweMarathon | Coming soon | 31.9% | Coming soon |

### Reasoning

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: 26.3 (Unranked · 3 rankable rows)
- Hy4 preview public-lane score: Coming soon

| Benchmark | DeepSeek V3 | Hy4 preview | Winner |
|-----------|-----------|-----------|--------|
| LongBench v2 | 48.7% | Coming soon | Coming soon |
| CritPt | Coming soon | 16.9% | Coming soon |

### Multimodal & Grounded

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: Coming soon
- Hy4 preview public-lane score: 88.3 (Unranked · 1 rankable row)

| Benchmark | DeepSeek V3 | Hy4 preview | Winner |
|-----------|-----------|-----------|--------|
| OfficeQA Pro | Coming soon | 66.2% | Coming soon |

### Knowledge

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: 32 (Supported · #142/180)
- Hy4 preview public-lane score: 63.6 (Estimated · #39/180)

| Benchmark | DeepSeek V3 | Hy4 preview | Winner |
|-----------|-----------|-----------|--------|
| GPQA | 59.1% | 92.3% | Hy4 preview |
| MMLU-Pro | 75.9% | Coming soon | Coming soon |
| SimpleQA | 24.9% | Coming soon | Coming soon |
| GPQA-D | Coming soon | 92.3% | Coming soon |
| HLE | Coming soon | 55.4% | Coming soon |
| HLE w/o tools | Coming soon | 43.4% | Coming soon |

### Instruction Following

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: 39.6 (#104/127)
- Hy4 preview public-lane score: Coming soon

| Benchmark | DeepSeek V3 | Hy4 preview | Winner |
|-----------|-----------|-----------|--------|
| IFEval | 86.1% | Coming soon | Coming soon |

### Mathematics

- Winner: Insufficient shared evidence
- DeepSeek V3 public-lane score: Coming soon
- Hy4 preview public-lane score: Coming soon

| Benchmark | DeepSeek V3 | Hy4 preview | Winner |
|-----------|-----------|-----------|--------|
| AIME 2024 | 39.2% | Coming soon | Coming soon |
| MATH-500 | 90.2% | Coming soon | Coming soon |
| Apex | Coming soon | 74.2% | Coming soon |

## FAQ

### Is there enough evidence to pick between DeepSeek V3 and Hy4 preview?

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 0324](/compare/deepseek-v3-vs-deepseek-v3-0324)
- [Hy4 preview vs DeepSeek V3 0324](/compare/deepseek-v3-0324-vs-hy4-preview)
- [DeepSeek V3 vs Claude Opus 5.5](/compare/claude-opus-5-5-vs-deepseek-v3)
- [Hy4 preview vs Claude Opus 5.5](/compare/claude-opus-5-5-vs-hy4-preview)

## Explore More

- [DeepSeek V3 profile](/models/deepseek-v3)
- [Hy4 preview profile](/models/hy4-preview)
- [Compare Pricing](/llm-pricing)
- [Alternative Finder](/tools/alternative-finder)
- [LLM Selector](/tools/llm-selector)
- [Overall Rankings](/best/overall)