Agentic
Directional only- DeepSeek V4 Pro (High)
- 70.6
- Hy3 Preview
- 54.4
- Weighted basis
- 2 vs 1 rows
- Reading
- Directional only
Model comparison
DeepSeek V4 Pro (High) vs Hy3 Preview
Updated August 7, 2026. Public scores include evidence status and uncertainty. They are not guarantees for a specific workload.
DeepSeek V4 Pro (High) has the higher public score estimate, 55.46 versus 42.8, but the 90% score intervals overlap. Treat that as a lead, not a settled winner.
5 results are shared. Category rows based on different benchmark sets are marked directional and do not name a winner.
Which one for your work
Recommendations appear only when a shared evidence basis or an explicit operating constraint supports the call. Secondary and unsupported use cases stay disclosed below the initial list.
Long documents
Prompts that approach the documented context limit
DeepSeek V4 Pro (High)
DeepSeek V4 Pro (High) has the larger documented context window.
Confidence: documented
Show secondary and unsupported calls
Coding work
Code generation, repair, and software-engineering tasks
Not enough matched evidence
The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.
Confidence: limited
Agentic work
Tool use, computer use, and multi-step task completion
Not enough matched evidence
The category averages use different weighted benchmark sets, so they are directional rather than like-for-like.
Confidence: limited
Chat turn cost
1K fresh input + 500 output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: listed-rates
Cache-heavy agent loop cost
200K cached + 20K fresh input + 10K output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: listed-rates
Repository review cost
50K fresh input + 3K output tokens
Not enough matched evidence
A complete comparable API-rate estimate is not available for both models.
Confidence: listed-rates
What is actually comparable
Shared results can support a head-to-head reading. Results present for only one model describe coverage, not superiority.
18 DeepSeek V4 Pro (High) only
5 Shared
1 Hy3 Preview only
- Shared results
- 5
- DeepSeek V4 Pro (High) only
- 18
- Hy3 Preview only
- 1
- Like-for-like categories
- 0 / 8
3 categories use different evidence sets. Those rows remain visible for coverage context but do not name a winner.
Category results, on a stated basis
Each row states whether both averages use the same weighted benchmark set. Directional and not-comparable rows remain visible, but they never receive a winner in this template.
Coding
Directional only- DeepSeek V4 Pro (High)
- 69.8
- Hy3 Preview
- 74.4
- Weighted basis
- 2 vs 1 rows
- Reading
- Directional only
Knowledge
Directional only- DeepSeek V4 Pro (High)
- 57.0
- Hy3 Preview
- 87.2
- Weighted basis
- 4 vs 1 rows
- Reading
- Directional only
Reasoning
Not comparable- DeepSeek V4 Pro (High)
- Not measured
- Hy3 Preview
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
Math
Not comparable- DeepSeek V4 Pro (High)
- 94.0
- Hy3 Preview
- Not measured
- Weighted basis
- 1 vs 0 rows
- Reading
- Not comparable
Multilingual
Not comparable- DeepSeek V4 Pro (High)
- Not measured
- Hy3 Preview
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
Multimodal
Not comparable- DeepSeek V4 Pro (High)
- Not measured
- Hy3 Preview
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
Instruction following
Not comparable- DeepSeek V4 Pro (High)
- Not measured
- Hy3 Preview
- Not measured
- Weighted basis
- 0 vs 0 rows
- Reading
- Not comparable
| Category | DeepSeek V4 Pro (High) | Hy3 Preview | Weighted basis | Reading |
|---|---|---|---|---|
| Agentic | 70.6 | 54.4 | Directional only2 vs 1 rows | Directional only |
| Coding | 69.8 | 74.4 | Directional only2 vs 1 rows | Directional only |
| Knowledge | 57.0 | 87.2 | Directional only4 vs 1 rows | Directional only |
| Reasoning | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
| Math | 94.0 | Not measured | Not comparable1 vs 0 rows | Not comparable |
| Multilingual | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
| Multimodal | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
| Instruction following | Not measured | Not measured | Not comparable0 vs 0 rows | Not comparable |
Shape of the matched evidence
Only shared public evidence is shown. Sparse evidence stays a ruled list rather than being closed into a radar shape.
Too few matched category axes support a radar. The ruled list below shows only shared benchmark results; positions use each benchmark’s normalized display scale when available.
Terminal-Bench 2.0
Agentic
DeepSeek V4 Pro (High): 63.3%Hy3 Preview: 54.4%Normalized gap 8.9DeepSeek V4 Pro (High) source Hy3 Preview sourceSWE-bench Verified
Coding
DeepSeek V4 Pro (High): 79.4%Hy3 Preview: 74.4%Normalized gap 5.0DeepSeek V4 Pro (High) source Hy3 Preview sourceGPQA
Knowledge
DeepSeek V4 Pro (High): 89.1%Hy3 Preview: 87.2%Normalized gap 1.9DeepSeek V4 Pro (High) source Hy3 Preview source
What each workload costs
Three fixed token mixes turn per-token rates into comparable decisions. Each scenario states context fit and whether cached input had to fall back to the published list-input rate.
Chat turn
1K fresh input + 500 output tokens
- DeepSeek V4 Pro (High)
- $0.00087
- Fits in one request
- Hy3 Preview
- Self-hosted; infrastructure cost varies
- Fits in one request
Hy3 Preview has no comparable published API token rate.
Repository review
50K fresh input + 3K output tokens
- DeepSeek V4 Pro (High)
- $0.02436
- Fits in one request
- Hy3 Preview
- Self-hosted; infrastructure cost varies
- Fits in one request
Hy3 Preview has no comparable published API token rate.
Cache-heavy agent loop
200K cached + 20K fresh input + 10K output tokens
- DeepSeek V4 Pro (High)
- $0.01812
- Fits in one request
- Hy3 Preview
- Self-hosted; infrastructure cost varies
- Fits in one request
- Cached-input rate unavailable
Hy3 Preview has no comparable published API token rate.
Specification differences
Sourced differences are shown directly. Missing facts stay explicit instead of being inferred from a model name or family.
SpecificationDeepSeek V4 Pro (High)Hy3 Preview
Context window
Maximum documented context; output-token limits may be lower.
DeepSeek V4 Pro (High)
Hy3 Preview
256K
API model ID
DeepSeek V4 Pro (High)
deepseek-v4-pro
DeepSeek models and pricingHy3 Preview
Not sourced
Cached-input rate
A missing cached-input rate falls back to the listed input rate only in the stated workload estimate.
DeepSeek V4 Pro (High)
$0.003625 per 1M cached input tokens
Hy3 Preview
No comparable hosted API rate
Documented inputs
DeepSeek V4 Pro (High)
Hy3 Preview
Not sourced
Documented outputs
DeepSeek V4 Pro (High)
Hy3 Preview
Not sourced
Provider availability
DeepSeek V4 Pro (High)
Preview · DeepSeek API, open weights
DeepSeek V4 Flash 0731 updateHy3 Preview
Not sourced
Reasoning profile
DeepSeek V4 Pro (High)
Reasoning
Hy3 Preview
Reasoning
Weight access
DeepSeek V4 Pro (High)
Open Weight
Hy3 Preview
Open Weight
License
DeepSeek V4 Pro (High)
Open Weight
Hy3 Preview
Open Weight
Release date
DeepSeek V4 Pro (High)
2026-04-24
Hy3 Preview
2026-04-23
If you already use one of these models
- Deployment change
- The models list different providers, so authentication, endpoint behavior, limits, and feature support may change.
- Quality signal
- DeepSeek V4 Pro (High) has the higher public score estimate, 55.46 versus 42.8, but the 90% score intervals overlap.
- Workload cost
- A complete comparable API-rate estimate is not available for both models.
- Context tradeoff
- DeepSeek V4 Pro (High) has the larger documented window (1M).
Run the same representative tasks against both endpoints before changing production traffic.
Benchmark evidence
The full public result ledger is available for audit without forcing a wide desktop table onto a phone.
Browse raw public benchmark evidence24 rows
Agentic
Terminal-Bench 2.0
DeepSeek V4 Pro (High)63.3%SourceHy3 Preview54.4%SourceDeepSeek V4 Pro (High) leads this result
BrowseComp
DeepSeek V4 Pro (High)80.4%SourceHy3 Preview—Not directly comparable
HLE w/ tools
DeepSeek V4 Pro (High)44.7%SourceHy3 Preview—Not directly comparable
MCP Atlas
DeepSeek V4 Pro (High)74.2%SourceHy3 Preview—Not directly comparable
Toolathlon
DeepSeek V4 Pro (High)49%SourceHy3 Preview—Not directly comparable
Gert Labs
DeepSeek V4 Pro (High)—Hy3 Preview36.91%SourceNot directly comparable
Coding
LiveCodeBench Pass@1-COT
DeepSeek V4 Pro (High)89.8%SourceHy3 Preview—Not directly comparable
Codeforces
DeepSeek V4 Pro (High)2919.0SourceHy3 Preview—Not directly comparable
SWE-bench Verified
DeepSeek V4 Pro (High)79.4%SourceHy3 Preview74.4%SourceDeepSeek V4 Pro (High) leads this result
SWE-bench Pro
DeepSeek V4 Pro (High)54.4%SourceHy3 Preview—Not directly comparable
SWE Multilingual
DeepSeek V4 Pro (High)74.1%SourceHy3 Preview—Not directly comparable
Terminal-Bench 2.0
DeepSeek V4 Pro (High)63.3%SourceHy3 Preview54.4%SourceDeepSeek V4 Pro (High) leads this result
Reasoning
Knowledge
MMLU-Pro
DeepSeek V4 Pro (High)87.1%SourceHy3 Preview—Not directly comparable
SimpleQA
DeepSeek V4 Pro (High)46.2%SourceHy3 Preview—Not directly comparable
Chinese-SimpleQA
DeepSeek V4 Pro (High)77.7%SourceHy3 Preview—Not directly comparable
GPQA
DeepSeek V4 Pro (High)89.1%SourceHy3 Preview87.2%SourceDeepSeek V4 Pro (High) leads this result
GPQA-D
DeepSeek V4 Pro (High)89.1%SourceHy3 Preview87.2%SourceDeepSeek V4 Pro (High) leads this result
HLE
DeepSeek V4 Pro (High)34.5%SourceHy3 Preview—Not directly comparable
Math
HMMT Feb 2026
DeepSeek V4 Pro (High)94.0%SourceHy3 Preview—Not directly comparable
IMOAnswerBench
DeepSeek V4 Pro (High)88.0%SourceHy3 Preview—Not directly comparable
Apex
DeepSeek V4 Pro (High)27.4%SourceHy3 Preview—Not directly comparable
Apex Shortlist
DeepSeek V4 Pro (High)85.5%SourceHy3 Preview—Not directly comparable
Frequently asked questions
Which is better, DeepSeek V4 Pro (High) or Hy3 Preview?
DeepSeek V4 Pro (High) has the higher public score estimate, 55.46 versus 42.8, but the 90% score intervals overlap. The higher estimate is not a decisive winner because the uncertainty ranges overlap.
Which is better for coding, DeepSeek V4 Pro (High) or Hy3 Preview?
The current coding averages use different weighted benchmark sets, so BenchLM does not name a winner from them. Read the shared benchmark rows directly and test the models on the same task set.
Which is better for agentic tasks, DeepSeek V4 Pro (High) or Hy3 Preview?
The current agentic tasks averages use different weighted benchmark sets, so BenchLM does not name a winner from them. Read the shared benchmark rows directly and test the models on the same task set.
Which costs less, DeepSeek V4 Pro (High) or Hy3 Preview?
Both models do not have comparable published API token rates, so this page does not name a universal price winner.
Which has the larger context window, DeepSeek V4 Pro (High) or Hy3 Preview?
DeepSeek V4 Pro (High) has the larger documented context window: 1M, compared with 256K.
Related comparisons
Last updated August 7, 2026
Watch DeepSeek V4 Pro (High) vs Hy3 Preview
One weekly email when material rank, price, or benchmark evidence changes make this matchup worth revisiting.
Read a sample issueJoin 2,000+ readers.