What is the latest DeepSeek-V4-Pro API version?
The current official API documentation lists DeepSeek-V4-Pro-0813. The API calling name remains deepseek-v4-pro, so existing integrations do not need to change the model parameter.
DE
DeepSeek V4 Pro
Release date: 2026-08-13Updated: 2026-08-13Knowledge cutoff: 2025-05Views: 16,951
Parameters
1.6T
Context length
1M
Multilingual
Supported
Reasoning ability
5/5
DeepSeek's current deepseek-v4-pro API version is DeepSeek-V4-Pro-0813, with the calling name unchanged. It is a text-only 1.6T-total/49B-active MoE model with a 1M-token context window and 384K maximum output, offering non-thinking, Think High, and Think Max modes plus the Responses API, Anthropic-compatible access, JSON output, and tool calls. Current prices per 1M tokens are $0.003625 cache-hit input, $0.435 cache-miss input, and $0.87 output. The public Hugging Face checkpoint remains labeled preview and is not represented as the 0813 API weights.
Data sourced primarily from official releases (GitHub, Hugging Face, papers), then benchmark leaderboards, then third-party evaluators. Learn about our data methodology
DeepSeek-V4-Pro
Model basics
Reasoning traces
Supported
Thinking modes
Thinking Level · Max (Default)Standard ModeThinking Level · High
Context length
1M tokens
Max output length
384K tokens
Model type
Reasoning model
Modality (in / out)
Text → Text
Release date
2026-08-13
Model file size
No data
MoE architecture
Yes
Total params / Active params
1.6T / 49B
Knowledge cutoff
2025-05
DeepSeek-V4-Pro
Open source & experience
DeepSeek-V4-Pro
Official resources
DataLearnerAI blog
N/A
DeepSeek-V4-Pro
API details
API speed
3/5
💡Default unit: $/1M tokens. If vendors use other units, follow their published pricing.
Standard
| Type | Condition | Input | Output |
|---|---|---|---|
| Text | - | $0.435/ 1M tokens | $0.870/ 1M tokens |
Cache PricingPrompt Cache
| Type | TTL | Write | Read |
|---|---|---|---|
| Text | - | — | $0.0036/ 1M tokens |
“—” means the modality is not billed in that direction, or the vendor has not published a price for it.
DeepSeek-V4-Pro
Benchmark Results
DeepSeek-V4-Pro currently shows benchmark results led by LiveCodeBench (1 / 128, score 93.50), MMLU Pro (11 / 134, score 87.50), SWE-bench Verified (11 / 116, score 80.60). This page also consolidates core specs, context limits, and API pricing so you can evaluate the model from benchmark results and deployment constraints together.
General Knowledge
9 evaluationsBenchmark / mode
Score
Rank/total
General Evaluation
3 evaluationsBenchmark / mode
Score
Rank/total
Coding and Software Engineer
16 evaluationsBenchmark / mode
Score
Rank/total
Writing and Creative Capabilities
1 evaluationsBenchmark / mode
Score
Rank/total
AI Agent - Information Search
2 evaluationsBenchmark / mode
Score
Rank/total
AI Agent - Tool Usage
6 evaluationsBenchmark / mode
Score
Rank/total
Text Embedding
2 evaluationsBenchmark / mode
Score
Rank/total
Math and Reasoning
5 evaluationsBenchmark / mode
Score
Rank/total
Productivity Knowledge
3 evaluationsBenchmark / mode
Score
Rank/total
Compare with other models
DeepSeek-V4-Pro
Publisher
DeepSeek-AI
View publisher details DeepSeek V4 Pro
Model Overview
Latest API version: DeepSeek-V4-Pro-0813
DeepSeek's current API documentation identifies deepseek-v4-pro as DeepSeek-V4-Pro-0813. The calling name remains deepseek-v4-pro, so DataLearner updates the existing model record rather than creating a duplicate 0813 entry. The API supports non-thinking and thinking operation, with thinking enabled by default.
Architecture and specifications
DeepSeek-V4-Pro is a text-input, text-output MoE model with 1.6T total parameters and 49B activated parameters, a 1M-token context window, and up to 384K output tokens. The V4 family combines Compressed Sparse Attention (CSA) and Heavily Compressed Attention (HCA), together with mHC connections and the Muon optimizer.
API features and pricing
The current API supports OpenAI-compatible Chat Completions, the Responses API, Anthropic-compatible access, JSON output, tool calls, prefix completion, and FIM completion in non-thinking mode. The documented concurrency limit is 500. The official pricing page lists $0.003625 per one million cache-hit input tokens, $0.435 per one million cache-miss input tokens, and $0.87 per one million output tokens. DeepSeek mentions a significant future price increase but gives no effective date or new rates, so no pending price is represented as active.
Open-weight and benchmark boundary
The official DeepSeek-V4-Pro Hugging Face page still describes its public checkpoint as a preview version and does not identify those weights as the 0813 API build. The current API and public checkpoint therefore should not be treated as identical without an explicit DeepSeek statement. DataLearner retains its existing structured V4 Pro/open-checkpoint benchmark rows; the current API documentation does not publish a separate official 0813 benchmark table, so older results are not relabeled as new 0813 scores.
DeepSeek-V4-Pro
FAQ
What input and output modalities does DeepSeek-V4-Pro support?
The official model accepts text and returns text. Image input is not supported.
What are the parameter count, context window, and maximum output?
DeepSeek-V4-Pro has 1.6T total parameters with 49B activated per token, a 1M-token context window, and up to 384K output tokens.
Which API features does DeepSeek-V4-Pro-0813 support?
It supports OpenAI-compatible Chat Completions, the Responses API, Anthropic-compatible access, JSON output, tool calls, prefix completion, and FIM in non-thinking mode. The documented concurrency limit is 500.
How much does the DeepSeek-V4-Pro-0813 API cost?
Current prices per 1M tokens are $0.003625 for cache-hit input, $0.435 for cache-miss input, and $0.87 for output. A future increase has been mentioned but is not active because no date or rates have been published.
Are DeepSeek-V4-Pro-0813 weights and dedicated benchmarks available?
The official Hugging Face checkpoint is still labeled preview and is not identified as the 0813 API build. The current API docs also contain no separate official 0813 benchmark table, so existing V4 Pro results are retained with that version boundary.
DataLearner on WeChat
Follow DataLearner on WeChat for AI model updates and research notes.
