Story perspectives

DeepSeek Launches Open-Source V4, 1M-Token Context, Beats Rivals

4/24/2026

Summary

Story summary

On April 24, 2026 DeepSeek released its V4 model in China as open-source, featuring a one-million-token context window.
V4-Pro (1.6 trillion parameters) and V4-Flash (284 billion) run on Huawei Ascend chips, with DeepSeek saying V4-Pro beats all open-source rivals and trails only Google’s Gemini-3.1-Pro.
V4-Pro costs $3.48 and V4-Flash $0.28 per million tokens, far below OpenAI’s $30 and Anthropic’s $25, as SMIC shares rose 10% while Minimax and Knowledge Atlas each fell over 9%.

History

Historical Context

In February 2019 OpenAI announced GPT-2, a 1.5-billion-parameter transformer that it initially withheld due to misuse concerns. The partial release and later open-source replications showed large language models could be reproduced and extended, igniting a wave of commercial LLM development and pricing models that shape today’s global AI market economics.

1 of 2

Story summary

On April 24, 2026 DeepSeek released its V4 model in China as open-source, featuring a one-million-token context window.
V4-Pro (1.6 trillion parameters) and V4-Flash (284 billion) run on Huawei Ascend chips, with DeepSeek saying V4-Pro beats all open-source rivals and trails only Google’s Gemini-3.1-Pro.
V4-Pro costs $3.48 and V4-Flash $0.28 per million tokens, far below OpenAI’s $30 and Anthropic’s $25, as SMIC shares rose 10% while Minimax and Knowledge Atlas each fell over 9%.

1 / 2