Is DeepSeek v4 Flash 0731 available with Zero Data Retention on Elosia?

DeepSeek v4 Flash 0731 runs on Elosia's standard backend; ZDR is available on many other models in the catalog.

How much does DeepSeek v4 Flash 0731 cost on Elosia?

DeepSeek v4 Flash 0731 is billed by usage at $0.08 per 1M input tokens and $0.18 per 1M output tokens, within your Elosia plan.

What is DeepSeek v4 Flash 0731's context window?

DeepSeek v4 Flash 0731 supports a context window of up to 1,048,576 tokens.

Is DeepSeek v4 Flash 0731 hosted in the EU?

Yes, DeepSeek v4 Flash 0731 is available on Elosia's EU-hosted (EdenAI) backend for enterprise organizations.

When should you choose DeepSeek v4 Flash 0731?

DeepSeek v4 Flash 0731 is a balanced model: a strong quality/cost trade-off for everyday production work.

DeepSeek v4 Flash 0731

DeepSeekBalanced

ThinkingTool UseStructured Output

About this model

DeepSeek V4 Flash 0731 is a sparse mixture-of-experts model from DeepSeek, with 13B active parameters out of 284B total. This re-post-trained revision is suited for coding, reasoning, and agent workflows....

Performance Tier

Balanced

DeepSeek v4 Flash 0731 is a balanced model from DeepSeek : strong performance at a reasonable price.

Strong cost-performance ratio. Reliable for most professional use cases without premium pricing.

Pricing

This model is included in Elosia plans

Eco

Minimal cost. Ideal for very high volume or simple tasks.

Type	per 1M tokens
Input (prompt)	$0.080
Output (completion)	$0.180
Cache read	$0.016

Capabilities

Context Length1.0M

Max Output Tokens384K

TokenizerDeepSeek

Inputtext

Outputtext

Release DateJuly 31, 2026

Benchmarks

General Intelligence

MMLU

Not reported

GPQA Diamond

90.8%

Mathematics

MATH-500

Not reported

Programming

HumanEval

Not reported

Reasoning

Humanity's Last Exam

38.6%

Agentic

Terminal-Bench 2.1

82.7%

Where does DeepSeek v4 Flash 0731 stand?

Compare its performance index against every other model.

View the performance leaderboard

Recommended Use Cases

CodingAnalysisResearchMathematics

Strengths

Mixture of experts with 13B active parameters per token, drawn from 256 routed experts plus one shared, six engaged at a time
Agentic and coding execution evaluated across four separate suites, covering terminal workflows, repository patches, security tasks and tool use
Speculative decoding built into the checkpoint itself: the DSpark module ships with the weights rather than being bolted on at serving time
MIT open weights that are genuinely self-hostable, with official vLLM and SGLang recipes, tool calling with parallel_tool_calls, a 1M-token context and 384K of output

Limitations

Weak factual recall: on questions whose answer it does not know, it answers anyway far more often than it abstains. The strength is on agentic and code, not on knowledge density
Official figures are not reproducible, the DeepSeek Harness not being published, so the announced results cannot be replayed in house. Verbose reasoning as well at maximum effort, which weighs on latency and on the token budget