0:00 / 0:00

DeepSeek V4.1 Flash: open weights, MIT, 552B params activating 8B/16B per token.

Name: DeepSeek V4.1 Flash: open weights, MIT, 552B params activating 8B/16B per token.
Uploaded: 2026-09-10T17:40:09.000Z
Duration: 1 min 7 s
Description: KV cache 890 bytes/token, ~1/4 of the previous Flash - the weights did NOT shrink. On Terminal-Bench 2.1 at maximum effort it scores 90.6 vs Opus 5.0's 89.1, but 30.0 and 31.2 on 3.0 and 4.0.

KV cache 890 bytes/token, ~1/4 of the previous Flash - the weights did NOT shrink. On Terminal-Bench 2.1 at maximum effort it scores 90.6 vs Opus 5.0's 89.1, but 30.0 and 31.2 on 3.0 and 4.0.

Andrew DariusSep 10

Keep scrolling for the next clip ↓All of deepseek Archive

0:00 / 0:00

4 DGX Sparks running DeepSeek V4.1 Flash

6 concurrent sessions, 170.73 tok/s coding throughput. 🔥 1,200 output tokens in 7.028 seconds. Warm single-stream: 💻 Code: 77.61 tok/s 🧮 Math: 74.08 🧠 Reasoning: 48.75 ✍️ Prose: 33.02 6-stream aggregate: 💻 Code: 170.73 tok/s 🔢

WëscheSep 1019

Keep scrolling for the next clip ↓All of deepseek Archive