Back to today’s briefing

Story perspectives

DeepSeek Launches Game-Changing Model, Slashes API Pricing

10/1/2025

28 3 Full Breakdown

Summary

Story summary

DeepSeek launched experimental model DeepSeek-V3.2-Exp on September 29, 2025, introducing Sparse Attention for long-context efficiency.
The model is open-sourced under the MIT License, enabling broad access and modification.
API pricing fell over 50% to $0.28 per million input tokens and $0.42 per million output tokens.
Analysts warn sparse attention may miss details; the release heightens competition with OpenAI and Alibaba.

1 of 1

Story summary

DeepSeek launched experimental model DeepSeek-V3.2-Exp on September 29, 2025, introducing Sparse Attention for long-context efficiency.
The model is open-sourced under the MIT License, enabling broad access and modification.
API pricing fell over 50% to $0.28 per million input tokens and $0.42 per million output tokens.
Analysts warn sparse attention may miss details; the release heightens competition with OpenAI and Alibaba.