DeepSeek V4 Pro 0813 vs Hy4 preview
Updated October 21, 2026. 22 benchmarks have a result for both. Coding is ranked on the Vibe Coding Index; every other board is shown as published.
Decision reading
Pick Hy4 preview for coding: it scores 61.6 on the Vibe Coding Index against 54.6 for DeepSeek V4 Pro 0813.
Pick DeepSeek V4 Pro 0813 when output tokens cost 1.3x less, or when what matters is Terminal-Bench 4 (Vals AI) or Vibe Code Bench (Vals AI), where it leads.
- Shared results
- 22
- DeepSeek V4 Pro 0813 only
- 56
- Hy4 preview only
- 2
- Like-for-like categories
- 5 of 9
Shared results can support a head-to-head reading. Results only one model has describe coverage, not superiority.
Which one for your work
Each card is decided by one number both models carry. Where only one has it, the card says so and makes no call.
Coding
Vibe Coding Index, four public coding benchmarks
Hy4 preview
Hy4 preview scores 61.6 against 54.6.
One benchmark behind each score
Intelligence
AA Intelligence Index
No call
Only DeepSeek V4 Pro 0813 has a AA Intelligence Index result, 36.0.
Directional only
Agentic work
AA Agentic Index
No call
Only DeepSeek V4 Pro 0813 has a AA Agentic Index result, 41.3.
Directional only
Reasoning
GPQA Diamond
No call
Only DeepSeek V4 Pro 0813 has a GPQA Diamond result, 92.4%.
Directional only
Knowledge
MMLU Pro
No call
Only DeepSeek V4 Pro 0813 has a MMLU Pro result, 87.0%.
Directional only
General
Vals Index
Hy4 preview
Hy4 preview leads, 53.2% against 53.0%.
Same board, both measured
Long documents
Listed context window
No call
Both take 1.05M tokens.
Listed by OpenRouter
Everyday cost
1K input + 500 output tokens
DeepSeek V4 Pro 0813
DeepSeek V4 Pro 0813 $0.0016, Hy4 preview $0.0021 per turn.
List prices
Cost per solved task
One SWE-bench Verified task, solved
No call
Only DeepSeek V4 Pro 0813 has the number, $0.10.
Directional only
At a glance
- Coding index
- DeepSeek V4 Pro 0813 54.6 from 1 benchmark, Hy4 preview 61.6 from 1 benchmark.
- Widest category
- Math, 25.0 points apart on average over 1 shared benchmark.
- Widest single gap
- ProofBench: DeepSeek V4 Pro 0813 50.0% against Hy4 preview 75.0%, 25.0 points apart on a 0 to 100 scale.
- Output price
- DeepSeek V4 Pro 0813 $1.98 against Hy4 preview $2.501 per million output tokens, 1.3x apart.
- Context window
- Both take 1.05M tokens.
Benchmark by benchmark
First the score: boards each model leads, per category, where both have a result. Then every board, a category to a fold; bars are each result on a 0 to 100 scale, so compare the two in a row, not one row with another.
| Category | DeepSeek V4 Pro 0813 | Bars | Hy4 preview | Ties |
|---|---|---|---|---|
| Coding | 3 | 6 | 1 tie | |
| Agentic work | 2 | 5 | ||
| Math | 0 | 1 | ||
| Knowledge | 0 | 2 | ||
| General | 0 | 2 | ||
| All shared boards | 5 | 16 | 1 tie |
CodingLike-for-likeFour of these boards feed the Vibe Coding Index36
| Benchmark | DeepSeek V4 Pro 0813 | Hy4 preview | Leader |
|---|---|---|---|
| LMArena WebDev | 1582 | 1631 | Hy4 preview |
| AA Coding Index | 68.8 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LiveCodeBench (Vals AI) | 87.5% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| SWE-bench Verified (Vals AI) | 96.4% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| Terminal-Bench 4 (Vals AI) | 14.1% | 8.1% | DeepSeek V4 Pro 0813 |
| Vibe Code Bench (Vals AI) | 82.3% | 77.5% | DeepSeek V4 Pro 0813 |
| Code migration (Vals AI) | 41.5% | 47.4% | Hy4 preview |
| ProgramBench (Vals AI) | 0.0% | 0.0% | Tie |
| IOI (Vals AI) | 51.6% | 59.3% | Hy4 preview |
| Terminal-Bench 2.1 (Vals AI) | 54.7% | 55.1% | Hy4 preview |
| Vibe Code Bench 1-100 (Vals AI) | 17.5% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| SkillsBench (Vals AI) | 53.8% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| CyberBench (Vals AI) | 72.9% | 65.4% | DeepSeek V4 Pro 0813 |
| SciCode | 51.0% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| FrontierCode | 28.5% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| WeirdML | 66.2% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| ALE-Bench | 1403.2 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: coding | 1469 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: software and IT | 1462 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena WebDev: React | 1588 | 1653 | Hy4 preview |
| LMArena WebDev: HTML | 1566 | 1620 | Hy4 preview |
| Neither model has a result on the other 17 boards here | |||
IntelligenceDirectional onlyPublic boards, shown as published, kept out of the index2 boards, none shared
| Benchmark | DeepSeek V4 Pro 0813 | Hy4 preview | Leader |
|---|---|---|---|
| AA Intelligence Index | 36.0 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| Epoch Capabilities Index | 155.6 | Not measured | Only DeepSeek V4 Pro 0813 measured |
Agentic workLike-for-likePublic boards, shown as published, kept out of the index25
| Benchmark | DeepSeek V4 Pro 0813 | Hy4 preview | Leader |
|---|---|---|---|
| AA Agentic Index | 41.3 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| SREBench (Vals AI) | 1.9% | 2.3% | Hy4 preview |
| Finance Agent (Vals AI) | 50.4% | 55.1% | Hy4 preview |
| Tax Agent Bench (Vals AI) | 26.8% | 32.6% | Hy4 preview |
| Legal Research Bench (Vals AI) | 40.9% | 45.2% | Hy4 preview |
| Harvey's Legal Agent Benchmark (Vals AI) | 10.8% | 8.8% | DeepSeek V4 Pro 0813 |
| Excel Modeling Benchmark (Vals AI) | 52.8% | 58.1% | Hy4 preview |
| BioMysteryBench (Vals AI) | Not measured | 69.3% | Only Hy4 preview measured |
| Terminal-Bench-Science (Vals AI) | 4.3% | 1.4% | DeepSeek V4 Pro 0813 |
| APEX-Agents | 47.3% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| Neither model has a result on the other 29 boards here | |||
ReasoningDirectional onlyPublic boards, shown as published, kept out of the index11 boards, none shared
| Benchmark | DeepSeek V4 Pro 0813 | Hy4 preview | Leader |
|---|---|---|---|
| GPQA Diamond | 92.4% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| GPQA Diamond (Artificial Analysis run) | 92.8% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| Humanity's Last Exam (Artificial Analysis run) | 41.0% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| GPQA Diamond (Epoch AI run) | 91.7% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| ARC-AGI | 90.5% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| ARC-AGI-2 | 61.3% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| Chess puzzles | 46.0% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| Mystery game puzzles | 43.0% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: hard prompts | 1458 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: expert | 1459 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: hard prompts in English | 1463 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| Neither model has a result on the other 10 boards here | |||
MathLike-for-likePublic boards, shown as published, kept out of the index01
| Benchmark | DeepSeek V4 Pro 0813 | Hy4 preview | Leader |
|---|---|---|---|
| OTIS Mock AIME | 98.6% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| FrontierMath, tiers 1 to 3 | 64.6% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| FrontierMath, tier 4 | 26.8% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| ProofBench | 50.0% | 75.0% | Hy4 preview |
| LMArena Text: math | 1441 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: mathematical work | 1443 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| Neither model has a result on the other 7 boards here | |||
KnowledgeLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | DeepSeek V4 Pro 0813 | Hy4 preview | Leader |
|---|---|---|---|
| MMLU Pro | 87.0% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LegalBench (Vals AI) | 82.4% | 83.8% | Hy4 preview |
| CorpFin (Vals AI) | 65.4% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| TaxEval (Vals AI) | 73.1% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| MedCode (Vals AI) | 42.5% | 43.3% | Hy4 preview |
| SimpleQA Verified | 52.9% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: science | 1457 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: medicine and healthcare | 1446 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: legal and government | 1459 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: business and finance | 1442 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| Neither model has a result on the other 7 boards here | |||
Web designNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | DeepSeek V4 Pro 0813 | Hy4 preview | Leader |
|---|---|---|---|
| Neither model has a result on 9 boards here | |||
GeneralLike-for-likePublic boards, shown as published, kept out of the index02
| Benchmark | DeepSeek V4 Pro 0813 | Hy4 preview | Leader |
|---|---|---|---|
| Vals Index | 53.0% | 53.2% | Hy4 preview |
| Public Benefits Bench (Vals AI) | Not measured | 68.6% | Only Hy4 preview measured |
| MedScribe (Vals AI) | 80.2% | 83.6% | Hy4 preview |
| Long context reasoning (AA-LCR) | 80.3% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| DTBench | 93.9% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMCA | 45.5% | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text | 1446 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: instruction following | 1448 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: creative writing | 1439 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: English | 1457 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: non-English | 1431 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: Chinese | 1481 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: Korean | 1449 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: French | 1446 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: Spanish | 1435 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: Russian | 1448 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: multi-turn | 1454 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: longer queries | 1458 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: writing and language | 1443 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| LMArena Text: entertainment and media | 1424 | Not measured | Only DeepSeek V4 Pro 0813 measured |
| Neither model has a result on the other 26 boards here | |||
Image generationNot comparablePublic boards, shown as published, kept out of the index0 boards, none shared
| Benchmark | DeepSeek V4 Pro 0813 | Hy4 preview | Leader |
|---|---|---|---|
| Neither model has a result on 5 boards here | |||
What each workload costs
Three fixed token mixes at each model's listed price, so per-token rates become a bill.
Chat turn
1K input + 500 output tokens
DeepSeek V4 Pro 0813 costs less
Repository review
50K input + 3K output tokens
DeepSeek V4 Pro 0813 costs less
Agent loop
200K input + 10K output tokens
DeepSeek V4 Pro 0813 costs less
Listed OpenRouter rates, without cache or batch discounts. A workload that does not fit a model's context window would need splitting.
Specification differences
The leading value in a row is marked; a row with a missing value names no leader. Our two rows are marked in red.
| Specification | DeepSeek V4 Pro 0813 | Tencent Hy4 preview |
|---|---|---|
| Maker | DeepSeek | Tencent |
| Weights | Open weights | Open weights |
| Released | August 12, 2026 | August 28, 2026 |
| Context window | 1.05M | 1.05M |
| Input $/Mtok | $0.66 | $0.834 |
| Output $/Mtok | $1.98 | $2.501 |
| Coding index | 54.6 | 61.6 |
| Independent measurements behind it | 1 | 1 |
| Cost per solved task | $0.10 | Not computable |
| Source | OpenRouter listing | OpenRouter listing |
Questions
Is DeepSeek V4 Pro 0813 better than Hy4 preview for coding?
Pick Hy4 preview for coding: it scores 61.6 on the Vibe Coding Index against 54.6 for DeepSeek V4 Pro 0813. Pick DeepSeek V4 Pro 0813 when output tokens cost 1.3x less, or when what matters is Terminal-Bench 4 (Vals AI) or Vibe Code Bench (Vals AI), where it leads.
Which is cheaper, DeepSeek V4 Pro 0813 or Hy4 preview?
DeepSeek V4 Pro 0813 costs $0.66 per million input tokens and $1.98 per million output tokens; Hy4 preview costs $0.834 and $2.501. Cost per solved task is known for DeepSeek V4 Pro 0813 only.
Which has the bigger context window, DeepSeek V4 Pro 0813 or Hy4 preview?
Neither: both take 1.05M tokens.
How many benchmarks is the DeepSeek V4 Pro 0813 vs Hy4 preview comparison based on?
22 benchmarks have a result for both models as of October 21, 2026. DeepSeek V4 Pro 0813's coding index rests on 1 independent benchmark and Hy4 preview's on 1. Where only one model has a result, the page says so instead of naming a winner.