Minaco

DeepSeek V4 Pro (0813)

This is the pinned August 13 release of DeepSeek V4 Pro, offered separately from Flash.

When to try it.

Compare Pro and Flash on the same repository task when deciding whether the additional output cost improves the result.

What to weigh

Do not infer that the Pro name guarantees a better answer for every task. Keep the same inputs and acceptance criteria when comparing.

A first task

Analyze this difficult defect, identify a reproducible cause and propose a fix with a verification plan.

This is a suggested evaluation task. Check the result against your own requirements; it is not a benchmark or a guarantee of performance.

Know the rate.

Currently published for DeepSeek Harness. Rates below are Minaco Credits per one million tokens.

Input
222
Cached input
194
Output
8,880

Cache writes: 222 Credits per million tokens.

Estimate a model call

Example: 10,000 uncached input tokens and 1,000 output tokens cost approximately 11.1 Credits before per-request rounding.

An agent turn may make several model calls. Context, tool results and billable reasoning tokens affect the total; enter cache-write tokens separately when applicable. The prompt total determines the rate for the whole request. This estimate is not a charge cap or a checkout quote. How Credits work.

Exact route: deepseek/deepseek-v4-pro-0813

Catalog checked 2026-11-17. Model access depends on your plan and current service availability.

Use it in your workspace.

  1. Prepare DeepSeek Harness

    Install Minaco on your Apple silicon Mac, connect your account and prepare the native agent in Agent settings.

  2. Choose the exact model

    Select DeepSeek Harness and a Minaco-credit account mode. If your plan offers DeepSeek V4 Pro (0813), choose it from the model selector. Desktop 0.13.14 or later reads the server’s model catalog.

  3. Give it a bounded task

    Open the project folder, provide the necessary context and review any approvals. Inspect the result and usage before expanding the scope.

When the next stage calls for a different agent, finish the turn and switch in the same conversation. Minaco prepares the visible working context for the next native session; hidden reasoning and credentials do not transfer.

How context handoff works →

Read the source.

DeepSeek model documentation describes the upstream model.

Editorial review: 2026-11-11. Task suggestions and comparisons are Minaco’s guidance, not vendor benchmark results. Minaco is an independent workspace; model API features do not imply that every feature is exposed by the native agent.

Compare nearby choices.

  • DeepSeek V4.1 Flash (1017) — This is the October 17 update of DeepSeek V4.1 Flash. DeepSeek’s release describes improvements in project creation, including interactive and visual applications.

Keep the work moving.

Your agents. Your workspace. One place to continue.

Download for Mac