AI Potluck
Product / UX / UI & API

DeepSeek Chat

DeepSeek

Chinese consumer chat app from DeepSeek (Hangzhou) on chat.deepseek.com plus iOS/Android apps, exposing DeepSeek V3/R1 models with thinking mode and web search. ~131.5M MAU in January 2026 (90% MoM surge), second only to ChatGPT in global AI DAU rankings, driven by free access to frontier-tier models.

Hosted consumer chat UI at chat.deepseek.com (web + iOS/Android apps), currently fronting DeepSeek-V4 Preview (stronger agent + reasoning). Verified live June 2026. Note: the underlying DeepSeek-V3/V4 model weights are open (MIT) and scored separately in base_pretrained; this record scores the CHAT PRODUCT/UI itself, which is a proprietary free hosted service, not open source software.

Openness

1 high confidence
1.0
client
proprietary(no public source for chat.deepseek.com app)
backend
hosted-SaaS
note
underlying models open(MIT) but the chat application/service is closed

The DeepSeek chat surface is a proprietary hosted service; openness of the V3/V4 weights it serves is a separate (model-layer) score. Scored as closed counterpart per ui_api recipe.

  • https://www.deepseek.com/en/ recorded 2026-06-04

    Free hosted chat at chat.deepseek.com serving DeepSeek-V4 Preview; no source release for the chat app itself

Adoption

5 medium confidence
5.0

DeepSeek's consumer app was one of the most-downloaded global AI apps after the early-2025 surge; the hosted chat surface remains a top-tier global assistant. No clean current MAU figure published on the vendor's own page, so confidence medium; reach >10M is well-supported by app-store ranking history. Attributed to the chat surface, not the model SKU.

Capability

3 medium confidence
3.0

Capable single-vendor consumer chat UI with strong underlying model, but narrow on the ui_api feature axes (one provider, limited enterprise/RAG/plugin breadth) vs frontier multi-model chat UIs. Not a category frontier-definer.

  • https://www.deepseek.com/en/ recorded 2026-06-04

    Chat product features: file reading, long-context, V4 agent capabilities, single-vendor model

Unchanged since 2026-06-09 (last edited, not re-checked)