tweet2026-04-25c/SFdasUmt
๐๐ฒ๐ฒ๐ฝ๐ฆ๐ฒ๐ฒ๐ธ ๐ฉ๐ฐ ๐ท๐๐๐ ๐ฑ๐ฟ๐ผ๐ฝ๐ฝ๐ฒ๐ฑ ๐ณ๐ฟ๐ผ๐บ ๐๐ต๐ถ๐ป๐ฎ ๐ฎ๐ ๐ผ๐ฝ๐ฒ๐ป ๐๐ผ๐๐ฟ๐ฐ๐ฒ.
Output formatready
x-contentmd
831 chars ยท 208 tokensUTF-8 ยท LF
# ๐๐ฒ๐ฒ๐ฝ๐ฆ๐ฒ๐ฒ๐ธ ๐ฉ๐ฐ ๐ท๐๐๐ ๐ฑ๐ฟ๐ผ๐ฝ๐ฝ๐ฒ๐ฑ ๐ณ๐ฟ๐ผ๐บ ๐๐ต๐ถ๐ป๐ฎ ๐ฎ๐ ๐ผ๐ฝ๐ฒ๐ป ๐๐ผ๐๐ฟ๐ฐ๐ฒ.
ย
ย
---
ย
๐๐ฒ๐ฒ๐ฝ๐ฆ๐ฒ๐ฒ๐ธ ๐ฉ๐ฐ ๐ท๐๐๐ ๐ฑ๐ฟ๐ผ๐ฝ๐ฝ๐ฒ๐ฑ ๐ณ๐ฟ๐ผ๐บ ๐๐ต๐ถ๐ป๐ฎ ๐ฎ๐ ๐ผ๐ฝ๐ฒ๐ป ๐๐ผ๐๐ฟ๐ฐ๐ฒ.
ย
V4 Pro packs 1.6 trillion params, with 49 billion active.
ย
V4 Flash runs lighter at 284 billion params.
ย
Both ship with a 1 million token context window.
ย
It scored 57.9 on Simple QA, beating Claude Opus 4.6.
ย
Pro uses just 27% of the compute V3 needed.
ย
Try Expert and Instant mode free at http://chat.deepseek.com.
ย
Save this. Open source AI just keeps catching up.
ย

ย
ref ยท SFdasUmtxmdbot.com/c/SFdasUmt