Model Tracker

DeepSeek-V4-Flash

DeepSeek-V4-Flash is an AI model made by DeepSeek AI. DeepSeek AI released it on 24 April 2026. It is retired. DeepSeek AI retired it on 10 September 2026.

In inglese

DeepSeek AI

DeepSeek-V4-Flash is an AI model made by DeepSeek AI. DeepSeek AI released it on 24 April 2026. It is retired. DeepSeek AI retired it on 10 September 2026. It takes text and returns text. Its context window is 1,000,000 tokens. Its weights are open. Its licence is MIT License.

Released: 24 April 2026 — fe-static.deepseek.com, read 16 October 2026
Status: Retired — api-docs.deepseek.com, read 16 October 2026
Retired: 10 September 2026 — api-docs.deepseek.com, read 16 October 2026
Takes: Text — fe-static.deepseek.com, read 16 October 2026
Returns: Text — fe-static.deepseek.com, read 16 October 2026
Open weights: Yes — api-docs.deepseek.com, read 16 October 2026
Licence: MIT License — fe-static.deepseek.com, read 16 October 2026
Context window: 1,000,000 tokens — fe-static.deepseek.com, read 16 October 2026
Longest answer: Not confirmed: The maker's model card does not state a standard maximum output limit for DeepSeek-V4-Flash. The current model-list documentation gives a limit for DeepSeek-V4.1-Flash, the successor, not this retired model.
Knowledge cutoff: Not confirmed: No reviewed maker page or model card states a knowledge-cutoff date for DeepSeek-V4-Flash.
API price, input: Not confirmed: The current pricing page prices DeepSeek-V4.1-Flash, and the maker says the retired deepseek-v4-flash name is routed to V4.1-Flash. It does not state a current standard input price for the retired model itself.
API price, output: Not confirmed: The current pricing page prices DeepSeek-V4.1-Flash, and the maker says the retired deepseek-v4-flash name is routed to V4.1-Flash. It does not state a current standard output price for the retired model itself.
Where: Not confirmed: two readings disagree: api, app, weights and api, weights
Parameters: Not confirmed: The maker's pages conflict: the V4 release announcement states 284B total parameters for Flash, while the V4 model card states 285B. No single value can be confirmed from these pages.
Model card: https://fe-static.deepseek.com/chat/transparency/deepseek-V4-model-card-EN.pdf — fe-static.deepseek.com, read 16 October 2026

Benchmarks

Measured by others, not by the maker: each score with the name the measurer gives the model and the setting it ran, its day and its licence.

Arena

Text Arena, deepseek-v4-flash: 1,436 (1,432–1,440, 51,987 votes), published 15 October 2026
Code Arena (WebDev), DeepSeek-V4-Flash (High): 1,582 (1,573–1,592, 4,953 votes), published 18 October 2026

Arena’s leaderboard dataset (lmarena-ai/leaderboard-dataset on Hugging Face), CC BY 4.0: ratings from people’s votes between two anonymous models, as Arena publishes them. Data · CC BY 4.0

Epoch AI

Epoch Capabilities Index, DeepSeek-V4-Flash: 146.1 (143.7–148.1), as of 19 October 2026

Epoch AI, ‘Capabilities & benchmarking’, epoch.ai/benchmarks, CC BY 4.0: Epoch AI’s own runs and its Capabilities Index. Data · CC BY 4.0

Questions

When was DeepSeek-V4-Flash released?

DeepSeek AI released DeepSeek-V4-Flash on 24 April 2026.

Is DeepSeek-V4-Flash still available?

No: DeepSeek AI retired DeepSeek-V4-Flash on 10 September 2026.

What is the context window of DeepSeek-V4-Flash?

The context window of DeepSeek-V4-Flash is 1,000,000 tokens, as DeepSeek AI prints it.

Are the weights of DeepSeek-V4-Flash open?

Yes: DeepSeek AI publishes the weights of DeepSeek-V4-Flash for download.

What is the licence of DeepSeek-V4-Flash?

The licence of DeepSeek-V4-Flash is MIT License, as DeepSeek AI names it.

What does DeepSeek-V4-Flash take and return?

DeepSeek-V4-Flash takes text and returns text.

Sources

Change Log (read 16 October 2026)
DeepSeek V4 Preview Release (read 16 October 2026)
DeepSeek V4 - Model Card (read 16 October 2026)
Models & Pricing (read 16 October 2026)

All of the Model Tracker