Is DeepSeek v4 Pro 0813 available with Zero Data Retention on Elosia?

DeepSeek v4 Pro 0813 runs on Elosia's standard backend; ZDR is available on many other models in the catalog.

How much does DeepSeek v4 Pro 0813 cost on Elosia?

DeepSeek v4 Pro 0813 is billed by usage at $0.66 per 1M input tokens and $1.9800000000000002 per 1M output tokens, within your Elosia plan.

What is DeepSeek v4 Pro 0813's context window?

DeepSeek v4 Pro 0813 supports a context window of up to 1,024,000 tokens.

Is DeepSeek v4 Pro 0813 hosted in the EU?

Yes, DeepSeek v4 Pro 0813 is available on Elosia's EU-hosted (EdenAI) backend for enterprise organizations.

When should you choose DeepSeek v4 Pro 0813?

DeepSeek v4 Pro 0813 is a flagship model, best for the most demanding reasoning, agentic and long-context tasks.

DeepSeek v4 Pro 0813

DeepSeekFlagship

ThinkingTool UseStructured Output

About this model

DeepSeek V4 Pro 0813 is a large-scale mixture-of-experts model from DeepSeek. This is the GA release of DeepSeek V4 Pro.

Performance Tier

Flagship

DeepSeek v4 Pro 0813 is a flagship model from DeepSeek : the most capable in their lineup.

Best-in-class model from this provider. Highest performance across benchmarks, ideal for demanding tasks.

Pricing

This model is included in Elosia plans

Affordable

Low cost. Suitable for sustained use and high-volume interactions.

Type	per 1M tokens
Input (prompt)	$0.660
Output (completion)	$1.98
Cache read	$0.022

Capabilities

Context Length1.0M

Max Output Tokens384K

TokenizerDeepSeek

Inputtext

Outputtext

Release DateAugust 12, 2026

Benchmarks

General Intelligence

MMLU

Not reported

GPQA Diamond

92.8%

Mathematics

MATH-500

Not reported

Programming

HumanEval

Not reported

SWE-bench Verified

Not reported

Reasoning

IFEval

Not reported

Humanity's Last Exam

42.7%

Agentic

Terminal-Bench 2.1

87.9%

Where does DeepSeek v4 Pro 0813 stand?

Compare its performance index against every other model.

View the performance leaderboard

Recommended Use Cases

CodingAnalysisResearchMathematicsData Extraction

Strengths

Mixture of experts with 1.6T total parameters and 49B active per token, drawn from 384 routed experts plus one shared, six engaged at a time
Post-training focused on agentic work, evaluated across terminal workflows, repository generation, security tasks, tool use and business automation
Speculative decoding built into the checkpoint: the DSpark module ships with the weights and is enabled with a single flag in vLLM and SGLang
1M-token context served by compressed hybrid sparse attention, with up to 384K tokens of output at high and max reasoning effort
Three reasoning effort levels (low, high, max) plus a direct-answer mode, under MIT open weights

Limitations

Weak factual reliability: independent testing shows it almost always answers questions it does not know instead of abstaining
Official agentic results were obtained with the DeepSeek Harness, which is not published, so they cannot be replayed in house; independent runs of the same terminal benchmark land lower
Text only: no image, audio or video input
DeepSeek publishes no GPQA Diamond score for this release; the one shown comes from third-party testing