DeepSeek models available: current API models, open-weight releases, and legacy names explained

DeepSeek's model lineup is easier to understand when it is split into three layers: the model names that can be called through the hosted API, the open-weight releases that can be downloaded and self-hosted, and legacy names that remain useful only for migration or historical context. As of September 2026, the hosted API is centered on DeepSeek-V4.1-Flash through the model name deepseek-flash, while older identifiers such as deepseek-chat and deepseek-reasoner no longer represent the current API lineup.

This page is the Data Studios reference map for the overall DeepSeek model ecosystem. The separate API-model page covers callable identifiers and routing behavior in detail, while the open-weight page focuses on downloadable releases and local deployment.

··········

THE CURRENT DEEPSEEK MODEL MAP HAS THREE DIFFERENT LAYERS

A model family, an API model name, and a downloadable checkpoint are not interchangeable concepts. One API identifier can be rerouted to a newer model, while an older open-weight checkpoint can remain downloadable long after it disappears from the hosted API.

........

Layer	Current examples	What the name tells you
Hosted API	deepseek-flash; deepseek-v4-pro	The identifier sent to DeepSeek's API
Open-weight releases	V4.1 Flash; V4; V3.2; R1	A specific downloadable model/checkpoint family
Legacy API names	deepseek-v4-flash; deepseek-v4-flash-vision-exp; deepseek-chat; deepseek-reasoner	Compatibility or historical identifiers, not the clean current naming layer

........

··········

V4.1 FLASH IS NOW THE PRIMARY HOSTED MODEL

DeepSeek-V4.1-Flash launched in September 2026 and is called through the hosted API with model=deepseek-flash. The current documentation lists a 1 million token context window, a maximum output of 384K tokens, both thinking and non-thinking modes, tool calling, JSON output, Responses API support, Anthropic-compatible access, and native vision.

The legacy names deepseek-v4-flash and deepseek-v4-flash-vision-exp are still accepted for compatibility, but DeepSeek states that those older models are retired and their requests are served by V4.1 Flash.

··········

V4 PRO IS IN A TRANSITION STATE IN DEEPSEEK'S OWN DOCUMENTATION

DeepSeek's current pricing page says requests to deepseek-v4-pro are routed to V4.1 Flash until V4.1 Pro arrives, while the September 10 changelog says V4 Pro API service would continue after September 14. Because these two official pages are not fully aligned, Data Studios treats V4 Pro as a transition-state identifier rather than presenting one routing claim as settled fact.

··········

DEEPSEEK-CHAT AND DEEPSEEK-REASONER BELONG TO THE LEGACY API ERA

When V4 arrived in April 2026, DeepSeek announced that deepseek-chat and deepseek-reasoner would be discontinued on July 24, 2026. During their migration period they mapped to non-thinking and thinking modes of V4 Flash. They should therefore be treated as historical integration names, not as the model IDs developers should select for new deployments.

··········

OPEN-WEIGHT RELEASES CONTINUE TO EXIST OUTSIDE THE HOSTED API

DeepSeek continues to publish downloadable model weights. V4.1 Flash, V4 Flash, V3.2 and R1 remain relevant for self-hosting, research, fine-tuning and deployment through third-party inference stacks even when the hosted DeepSeek API uses a different or newer naming scheme.

The practical distinction is important: API availability can change by routing policy, while an open-weight release remains a specific checkpoint that an operator can deploy independently.

··········

DATA STUDIOS STATUS MATRIX: CURRENT, COMPATIBILITY, OR HISTORICAL

Data Studios classifies names by what a developer can infer from them today rather than by launch chronology alone.

........

Name	Status in September 2026	Best interpretation
deepseek-flash	Current	Primary hosted API route to V4.1 Flash
deepseek-v4-pro	Transition	Still documented, but official routing descriptions conflict
deepseek-v4-flash	Compatibility	Accepted alias routed to V4.1 Flash
deepseek-v4-flash-vision-exp	Compatibility	Accepted legacy vision alias routed to V4.1 Flash
deepseek-chat / deepseek-reasoner	Historical	Retired V4 migration names
V4.1 Flash open weights	Current open-weight release	Specific downloadable checkpoint

........

··········

CHOOSE THE MODEL LAYER BEFORE CHOOSING THE MODEL NAME

For a new API integration, start from the current hosted identifiers. For local deployment, choose a specific open-weight checkpoint based on hardware and framework support. For migration work, preserve legacy names only long enough to verify the new route and update configuration.

For API identifiers and routing behavior, see https://www.datastudios.org/post/deepseek-available-models-supported-api-models-version-differences-capabilities-comparison-and-a. For open-weight families and local deployment, see https://www.datastudios.org/post/deepseek-available-models-supported-model-families-version-differences-capability-comparison-and.

··········

THE DEEPSEEK LINEUP IS NOW A ROUTING SYSTEM AS WELL AS A MODEL CATALOG

The old idea of one product name mapping permanently to one model is increasingly inaccurate for DeepSeek. A current integration needs to track the callable API identifier, the model version behind it, and whether an older name is still a real model or only a compatibility alias.

··········

FOLLOW US FOR MORE

··········

DATA STUDIOS

··········

datastudios.org