DeepSeek model product
DeepSeek-V4-Pro
6T-parameter MoE model with 49B activated parameters and a 1M-token context window.
Updated Aug 12, 2026. Default version: DeepSeek V4 Pro
LLMBoard score73.8DeepSeek-V4-Pro-Max
Coverage80%21 benchmark families
Context window1MTokens
Official input price$0.435DeepSeek API
DeepSeek-V4-Pro Specifications
Technical details for the model's default version.
- Version
- DeepSeek V4 Pro
- Released
- Apr 24, 2026
- Knowledge cutoff
- May 1, 2025
- Parameters
- N/A
- Context window
- 1M
- Max output
- 384K
- Inputs
- text
- Outputs
- text
- Open weights
- Yes
- License
- Unspecified
DeepSeek-V4-Pro Capability Profile
This profile uses the model's current scored version. Arena ratings and prices are shown separately.
DeepSeek-V4-Pro Benchmark Results
Benchmark scores for DeepSeek-V4-Pro-Max.
| CorpusQA 1M | 62.0% | 01 | 2 | 100.0% | C | |
| CSimpleQA | 84.4% | 01 | 7 | 100.0% | C | |
| LiveCodeBench | 93.5% | 01 | 73 | 100.0% | C | |
| MathArena Apex | 90.2% | 01 | 7 | 100.0% | C | |
| MRCR 1M | 83.5% | 01 | 3 | 100.0% | C | |
| CodeForces | 100.0% | 02 | 16 | 93.3% | C | |
| HMMT Feb 26 | 95.2% | 02 | 11 | 90.0% | C | |
| IMO-AnswerBench | 89.8% | 05 | 19 | 77.8% | C | |
| MMLU-Pro | 87.5% | 06 | 129 | 96.1% | C | |
| SWE-bench Multilingual | 76.2% | 08 | 34 | 78.8% | C | |
| SWE-Bench Verified | 80.6% | 08 | 105 | 93.3% | C | |
| SimpleQA | 57.9% | 09 | 46 | 82.2% | C | |
| FrontierSWE | 29.0% | 12 | 15 | 21.4% | B | |
| Toolathlon | 51.8% | 12 | 31 | 63.3% | C | |
| FrontierCode 1.1 | 17.6% | 13 | 15 | 14.3% | B | |
| Terminal-Bench 2.0 | 67.9% | 14 | 49 | 72.9% | C | |
| MCP Atlas | 73.6% | 16 | 31 | 50.0% | C | |
| BrowseComp | 83.4% | 17 | 58 | 71.9% | C | |
| LiveBench | 73.6% | 20 | 38 | 48.6% | B | |
| Humanity's Last Exam | 48.2% | 23 | 93 | 76.1% | C | |
| GPQA | 90.1% | 24 | 234 | 90.1% | C | |
| SWE-Bench Pro | 55.4% | 30 | 45 | 34.1% | C |
DeepSeek-V4-Pro Arena Results
Preference and agent-evaluation results for the default version.
| agent steerability | overall | 24 | 0.0 | N/A | 32.4K | |
| agent | overall | 26 | -0.0 | N/A | 1.3M | |
| agent bash recovery steps | overall | 26 | 0.0 | N/A | 44.6K | |
| agent praise complaint | overall | 29 | -0.0 | N/A | 7.6K | |
| agent tool hallucination | overall | 32 | 0.0 | N/A | 1.2M | |
| agent task outcome explicit | overall | 33 | -0.0 | N/A | 19.6K | |
| text | overall | 37 | 1451.1 | 53,689 | N/A | |
| text factuality | overall | 42 | 1454.8 | 53,624 | N/A | |
| webdev | overall | 44 | 1445.1 | 12,982 | N/A | |
| text style control | overall | 49 | 1458.1 | 53,689 | N/A |
DeepSeek-V4-Pro Runtime Performance
Provider-specific output speed and catalog latency for DeepSeek V4 Pro. Runtime does not affect the capability score.
No runtime data
Browse runtime rankingsNo provider-specific speed or latency record is linked to the default version yet.
Output Speed is generated output tokens received per second. Catalog latency is reported separately from observed provider TTFT.
DeepSeek-V4-Pro Pricing
Official vendor API pricing appears first, followed by individual provider offers.
- Official API
- $0.435 input, $0.87 output per 1M
- Official provider
- DeepSeek
- Lowest third-party
- From $0.435 input, $0.87 output per 1M via Nvidia
- Tracked offerings
- 30
| UnoRouter | deepseek-v4-pro:free | global | N/A | N/A | 1M | |
| Nvidia | deepseek-ai/deepseek-v4-pro | global | $0.435 | $0.87 | 1M | |
| Pioneer | deepseek-ai/DeepSeek-V4-Pro | global | $0.435 | $0.87 | 1M | |
| ZenMux | deepseek/deepseek-v4-pro | global | $0.435 | $0.87 | 1M | |
| Merge Gateway | deepseek/deepseek-v4-pro | global | $0.435 | $0.87 | 1M | |
| Hugging Face | deepseek-ai/DeepSeek-V4-Pro | global | $0.435 | $0.87 | 1M | |
| Ofox | deepseek/deepseek-v4-pro | global | $0.45 | $0.88 | 1M | |
| CrossModel | deepseek/deepseek-v4-pro | global | $0.47 | $0.94 | 1M | |
| OrcaRouter | deepseek/deepseek-v4-pro | global | $0.56 | $1.12 | 1M | |
| OpenRouter | deepseek/deepseek-v4-pro | global | $0.6317 | $1.26 | 1M | |
| ai& | deepseek-ai/deepseek-v4-pro | global | $1 | $2.5 | 1M | |
| NanoGPT | deepseek/deepseek-v4-pro | global | $1.1 | $2.2 | 1M | |
| Weights & Biases | deepseek-ai/DeepSeek-V4-Pro | global | $1.15 | $2.55 | 1M | |
| Deep Infra | deepseek-ai/DeepSeek-V4-Pro | global | $1.3 | $2.6 | 1M | |
| GMI Cloud | deepseek-ai/DeepSeek-V4-Pro | global | $1.39 | $2.78 | 1M | |
| Kilo Gateway | deepseek/deepseek-v4-pro | global | $1.6 | $3.2 | 1M | |
| NovitaAI | deepseek/deepseek-v4-pro | global | $1.6 | $3.2 | 1M | |
| Impossibl | deepseek/deepseek-v4-pro | global | $1.74 | $3.48 | 1M | |
| Fireworks AI | accounts/fireworks/models/deepseek-v4-pro | global | $1.74 | $3.48 | 1M | |
| Baseten | deepseek-ai/DeepSeek-V4-Pro | global | $1.74 | $3.48 | 262.1K | |
| FastRouter | deepseek/deepseek-v4-pro | global | $1.74 | $3.48 | 1M | |
| ClinePass | cline-pass/deepseek-v4-pro | global | $1.74 | $3.48 | 1M | |
| SiliconFlow | deepseek-ai/DeepSeek-V4-Pro | global | $1.74 | $3.48 | 1M | |
| Abacus | deepseek-ai/DeepSeek-V4-Pro | global | $1.74 | $3.48 | 1M | |
| Together AI | deepseek-ai/DeepSeek-V4-Pro | global | $1.74 | $3.48 | 512K | |
| HPC-AI | deepseek/deepseek-v4-pro | global | $1.74 | $3.48 | 1M | |
| Nebius Token Factory | deepseek-ai/DeepSeek-V4-Pro | global | $1.75 | $3.5 | 1M | |
| TensorX | deepseek/deepseek-v4-pro | global | $1.75 | $3.5 | 1M |
Official prices use only the vendor's configured official Provider and positive standard USD PAYG rates. Third-party offers remain explicitly labeled.
DeepSeek-V4-Pro Versions
Available versions of this model. The score column identifies the version used in the overall ranking.
| DeepSeek V4 Pro | N/A | N/A | 1M | 384K | Yes | Unspecified | |
| DeepSeek-V4-Pro-Max | 73.8 | 1.6T | 1M | 393.2K | No | MIT |
What is DeepSeek-V4-Pro?
Key information about DeepSeek-V4-Pro and its available data.
6T-parameter MoE model with 49B activated parameters and a 1M-token context window. 2 at 1M-token context. The model also incorporates Manifold-Constrained Hyper-Connections (mHC) for stable signal propagation and is trained with the Muon optimizer for faster convergence.
Pre-trained on more than 32T tokens, V4-Pro-Max significantly advances open-source knowledge capabilities, achieves top-tier performance in coding benchmarks, and bridges the gap with leading closed-source models on reasoning and agentic tasks.
Data as of 2026-08-11.
DeepSeek-V4-Pro vs nearby models
Open a comparison with the three ranked models immediately above and below this model.
Models similar to DeepSeek-V4-Pro
Recommendations prioritize the same model type and family, then the closest LLMBoard score.
DE
DeepSeek-V4-Flash
DeepSeek
DE
DeepSeek-V3.2
DeepSeek
DE
DeepSeek-R1
DeepSeek
DE
DeepSeek-V3.1
DeepSeek
DE
DeepSeek-V3
DeepSeek
DE
DeepSeek-R1-Distill-Qwen
DeepSeek
FAQ
Common questions about DeepSeek-V4-Pro.
When was DeepSeek-V4-Pro released?
DeepSeek-V4-Pro's default version was released on Apr 24, 2026.
How much does DeepSeek-V4-Pro cost?
DeepSeek-V4-Pro's official API price is $0.435 per million input tokens and $0.87 per million output tokens via DeepSeek. The lowest tracked third-party offer starts at $0.435 input and $0.87 output via Nvidia.
Who created DeepSeek-V4-Pro?
DeepSeek-V4-Pro was created by DeepSeek.
What is the context window for DeepSeek-V4-Pro?
The default version has a 1M token context window.
Is DeepSeek-V4-Pro open weight?
Yes. The default version is marked as open weight under an unspecified license.
How many API providers offer DeepSeek-V4-Pro?
30 provider offerings are linked to the default version.
What models should I compare DeepSeek-V4-Pro with?
Nearby ranked alternatives include GPT-5.6-Luna, GPT-5.5-Pro, Kimi K2.6.