DeepSeekDeepSeek V4 Pro and V4 Flash

The DeepSeek learning curve

Choose the V4 model and thinking mode explicitly, then preserve reasoning context correctly across tool calls.

Start with the promptReview the evidence

The mental model

What changes the result

“Model mode and conversation plumbing matter as much as the words in the prompt.”

Three levels

Climb the curve in this order

Do not add production scaffolding before you can describe and measure a good single-task result.

  1. 01

    Choose Pro, Flash, and mode

    Start with the correct capability and latency profile.

    • Use V4 Flash for faster, economical work and V4 Pro for capability-sensitive tasks.
    • Use non-thinking mode for routine transformations and thinking mode for difficult reasoning.
    • Use the current V4 model names instead of retired compatibility aliases.
    Ready to move on when: The selected model and mode are recorded with the eval result, latency, and cost.
  2. 02

    Make outputs machine-safe

    Produce predictable results for automation.

    • For JSON mode, set the response format, say “JSON” in the prompt, and provide an example.
    • Set enough output tokens to avoid truncating structured data.
    • Define tools with clear descriptions and schemas; use strict mode only with its documented beta constraints.
    Ready to move on when: The output parses, validates, and handles the documented empty-content edge case.
  3. 03

    Preserve thinking across tools

    Keep long agent loops valid and coherent.

    • Return reasoning content with assistant tool calls exactly as the API documents.
    • Do not send unsupported tool-choice controls in thinking-mode integrations.
    • Test long conversations because provider-compatible SDKs can differ in reasoning replay behavior.
    Ready to move on when: A multistep tool loop completes without a 400 error and preserves the evidence used for the final answer.

Starter template

Begin here, then earn every extra instruction

Replace the bracketed text, run it on a real task, and add guidance only when you can name the failure it fixes.

deepseek-starter.txt
Mode: [thinking for complex reasoning / non-thinking for fast routine work]

Task:
[One concrete outcome]

Available tools:
- [tool name]: [when it should be used]

Output:
- Format: [text or JSON]
- Required fields: [fields]
- Validation: [observable check]

If JSON is required, return JSON matching this example:
{"result":"...","evidence":["..."]}

Patterns that work

Explicit mode selection

Treat thinking versus non-thinking as a product decision tied to complexity, speed, and token use.

JSON with an example

Combine the API response format with the word JSON, a sample object, and adequate output space.

Careful compatibility plumbing

OpenAI-compatible or Anthropic-compatible syntax does not remove DeepSeek-specific conversation requirements.

Common failure modes

Retired aliases

An integration still calls deepseek-chat or deepseek-reasoner after their July 24, 2026 retirement.

Fix: Migrate to deepseek-v4-pro or deepseek-v4-flash and choose thinking mode explicitly.

Dropped reasoning content

A thinking-mode tool loop fails on the next request.

Fix: Preserve the documented reasoning content in conversation history alongside the assistant tool call.

JSON by instruction alone

The model returns prose, incomplete JSON, or an empty body.

Fix: Use JSON response format, include a JSON example, size max tokens, validate, and retry the documented edge case.

Current as of July 26, 2026

What the latest guidance changes

DeepSeek’s official change log says the legacy deepseek-chat and deepseek-reasoner aliases were discontinued on July 24, 2026. This guide therefore uses the V4 Pro and V4 Flash names and current thinking-mode requirements.

Evidence ledger

Read the sources behind the advice

Official provider documentation takes priority. Maintainer field notes are useful evidence, but they are labeled separately.

DeepSeek API change log

official

DeepSeek

V4 model names, release timing, alias retirement, and current API migration state.

DeepSeek V4 release

official

DeepSeek

V4 Pro versus Flash positioning, context, interfaces, and dual thinking modes.

JSON output

official

DeepSeek

JSON response-format requirements, examples, truncation, and empty-content behavior.

Tool calls

official

DeepSeek

Thinking-mode tools and strict function-schema behavior.

This guide is reviewed at least every 30 days. Automated checks flag unavailable sources and overdue reviews; substantive advice changes require human verification.