← Models

DeepSeek-Reasoner

DeepSeek · DeepSeek-R1

Deprecated DeepSeek reasoning alias that currently maps to V4 Flash thinking mode until July 2026.

Part of DeepSeek-R1 family
Type
language
Context
1M tokens
Max Output
384K tokens
Status
deprecated
Input
$0.14/1M tok
Output
$0.28/1M tok
API Access
Yes
License
proprietary
reasoning cost-efficient math coding compatibility deprecated
Released January 2025 · Updated May 16, 2026

Overview

Freshness note: Model capabilities, limits, and pricing can change quickly. This profile is a point-in-time snapshot last verified on May 16, 2026.

DeepSeek-Reasoner is no longer the route to choose for new DeepSeek API work. DeepSeek’s April 2026 changelog says the legacy deepseek-reasoner model name currently maps to DeepSeek V4 Flash in thinking mode and is scheduled to be discontinued on July 24, 2026.

This page is retained for teams auditing older integrations, prompt tests, and cost reports. For new reasoning-heavy DeepSeek workloads, use DeepSeek V4 Pro when quality matters most or V4 Flash thinking mode when cost and throughput matter more.

Capabilities

As a compatibility alias, DeepSeek-Reasoner now inherits the relevant behavior of V4 Flash thinking mode during the transition window. It remains useful for existing systems that cannot immediately change model names, but production owners should plan explicit migration before the July 2026 cutoff.

Technical Details

DeepSeek’s current pricing page says deepseek-reasoner corresponds to the thinking mode of deepseek-v4-flash for compatibility. The V4 family supports a 1M context length and 384K maximum output, so this page records those current compatibility limits while treating the model name itself as deprecated.

Operationally, do not use the alias as a long-term abstraction. Change integrations to deepseek-v4-flash with thinking enabled or to deepseek-v4-pro, then rerun evaluations because prompt behavior, output length, and cost assumptions may shift.

Pricing & Access

During the compatibility period, DeepSeek lists deepseek-reasoner as corresponding to V4 Flash thinking mode. V4 Flash pricing is:

  • Input cache hit: $0.0028 per 1M tokens
  • Input cache miss: $0.14 per 1M tokens
  • Output: $0.28 per 1M tokens

Repeated-context workloads benefit materially from caching, but the more important planning point is the discontinuation date: July 24, 2026.

Access:

  • DeepSeek API platform
  • Third-party gateways and cloud wrappers in some regions

For high-throughput deployments, cache strategy is a major optimization lever.

Best Use Cases

Use DeepSeek-Reasoner only as a migration concern. It is acceptable for short-lived compatibility work, but new applications should call V4 Flash or V4 Pro explicitly.

It is less ideal for any greenfield integration because the name is scheduled to disappear.

Comparisons

  • DeepSeek V4 Flash: The explicit replacement target for the compatibility alias.
  • DeepSeek V4 Pro: Better option for harder reasoning when cost is less important.
  • DeepSeek-R1: Still useful as an open-weight reasoning family, separate from the deprecated hosted alias.