Loading page
DeepSeek
DeepSeek V4 Pro
DeepSeek V4 Pro is a large language model from DeepSeek in the DeepSeek V4 family, offering a 1,000,000-token context window.
Context 1.0M
Input from
$0.748 / 1M tokens
across 10 providers
API Pricing
Cheapest on OpenRouter — 44% below avg| Provider | Input / 1M | Output / 1M | Cached / 1M |
|---|---|---|---|
| $0.748 | $1.50 | $0.062 | |
| $0.870 | $1.74 | $0.174 | |
| $0.910 | $2.72 | - | |
| $1.30 | $2.60 | $0.100 | |
| $1.32 | $3.96 | - | |
| $1.32 | $3.96 | $0.130 | |
| $1.49 | $2.97 | $0.743 | |
| $1.60 | $3.20 | $0.135 | |
| $1.74 | $3.48 | - | |
| $1.74 | $3.48 | - | |
| $1.75 | $3.50 | $0.440 |
Prices updated daily. Last check: Sep 6, 2026
DeepSeek V4 Pro pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Intelligence
24.5 / 100
Output Speed
59.1 t/s
Latency (TTFT)
1.1s
Reasoning & Knowledge
- GPQA Diamond71.7%
- Humanity's Last Exam8.2%
Agentic & Tool Use
- Terminal-Bench Hard36.4%
- τ²-bench91.2%
Instruction & Long Context
- IFBench45.8%
- Long-Context Reasoning53.0%
Benchmarks measured Sep 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- DeepSeek
- Family
- DeepSeek V4
- Context Window
- 1.0M
- Modalities
- Text
Capabilities
- Tool Calling
- No
- Open Source
- No
- Aliases
- deepseek-ai/DeepSeek-V4-Pro, DeepSeek-V4-Pro, deepseek-v4-pro
Strengths & Limitations
Strengths
- 1,000,000-token context window, enough to hold large codebases or document sets in a single request
- Measured output throughput of approximately 78 tokens per second (Artificial Analysis)
- Time to first token measured at roughly 1.2 seconds, suitable for interactive as well as batch use
- "Pro" tier within the DeepSeek V4 family, positioned above the family's standard configuration
- DeepSeek models are commonly served by multiple inference providers, allowing price and latency comparison
- Consistent alias naming (deepseek-ai/DeepSeek-V4-Pro) makes provider-to-provider routing straightforward
Limitations
- Very long prompts consume large token volumes, so effective cost scales quickly with context usage regardless of the per-token rate
- Throughput near 78 tokens per second is mid-range, so very long generations take noticeable wall-clock time
- Reported benchmark figures come from a single measurement source and vary by hosting provider
- We do not track verified modality, tool-calling, or licensing details for this model — confirm with DeepSeek's documentation
- Provider availability and configuration for newer DeepSeek releases can vary, so long-context limits may be capped below 1M by individual hosts
Key Features
•1,000,000-token context window
•Pro tier of the DeepSeek V4 model family
•Measured output speed of ~78.5 tokens per second (Artificial Analysis)
•Measured time to first token of ~1,200 ms
•Available under the deepseek-ai/DeepSeek-V4-Pro model identifier
•Multi-provider hosting for price and latency comparison
About DeepSeek V4 Pro
DeepSeek V4 Pro is a language model developed by DeepSeek and released as part of the DeepSeek V4 family. The "Pro" designation places it as the higher-capability variant within that generation, positioned above the family's standard configuration for workloads that need more headroom.
The most concrete specification we track for DeepSeek V4 Pro is its context window of 1,000,000 tokens, which is large enough to hold entire code repositories, long document collections, or extended multi-turn agent transcripts in a single request. Independent throughput measurements from Artificial Analysis put output generation at roughly 78 tokens per second with a time to first token of about 1.2 seconds; both figures depend heavily on which provider serves the model, so compare the hosts listed on this page rather than assuming a single number.
In practice, models in this position are used for long-context retrieval and summarization, repository-scale code work, and agentic pipelines where conversation history accumulates. Because DeepSeek models are commonly offered by several independent inference providers, DeepSeek V4 Pro can typically be evaluated across multiple hosts with different latency and throughput profiles. Beyond the fields listed above, we do not track additional verified specifications for this model — check DeepSeek's own documentation for modality support, tool-calling behavior, and licensing details before committing to an integration.
Common Use Cases
DeepSeek V4 Pro suits workloads where prompt length is the binding constraint: whole-repository code review and refactoring, analysis across long document sets such as contracts or research corpora, and log or transcript analysis that would otherwise require chunking and stitching. The million-token window also helps in agentic setups where tool output and conversation history accumulate over many turns and truncation would lose important state. At roughly 78 output tokens per second with a ~1.2 second time to first token, it is workable for interactive assistants, though workloads that emit very long outputs — full document drafts or large code generations — will be latency-bound and may be better run asynchronously. For short, high-volume classification or routing tasks, a smaller model will usually be more economical.
Frequently Asked Questions
How much does DeepSeek V4 Pro cost?
Pricing varies by provider and by pricing type — input versus output tokens, cached input, and batch or dedicated capacity are all priced differently. Because long-context requests can consume very large token counts, the effective cost depends more on your prompt sizes than on the headline rate. See the pricing table on this page for current per-provider rates.
What is DeepSeek V4 Pro best used for?
It is best suited to long-context work: repository-scale code analysis, summarization and question answering across large document collections, and multi-turn agent workflows where history accumulates. The 1,000,000-token context window is its clearest differentiator.
How large is the context window?
DeepSeek V4 Pro supports a context window of 1,000,000 tokens. Note that individual inference providers sometimes serve a model with a lower configured maximum, so check the provider's documentation if you plan to use the full window.
How fast is DeepSeek V4 Pro?
Artificial Analysis measured output generation at about 78.5 tokens per second with a time to first token of roughly 1,200 milliseconds. These figures depend on the hosting provider, request length, and load, so treat them as a reference point rather than a guarantee.
How does DeepSeek V4 Pro differ from other models in the DeepSeek V4 family?
The Pro designation marks it as the higher-capability variant within the DeepSeek V4 generation, intended for heavier workloads than the family's standard configuration. For a specification-by-specification comparison, check the individual model pages for each V4 variant we track.