← Back to Videos

DeepSeek V4.1 Flash Builds Tetris in 5.21 Seconds

Loading video

In a Fireworks Ultrafast demo, DeepSeek V4.1 Flash generated a playable Tetris in 5.21 seconds at 1,521 tok/s, versus 77.62 seconds for Claude Opus 5.5.

LLM DeepSeek Inference Speed Code Generation Fireworks AI Model Serving

Category: llm

Author: @FireworksAI_HQ

Date: 2026-10-21T00:00:00

Duration: 91.566s

Source: https://x.com/FireworksAI_HQ/status/2112973992097468795

DeepSeek V4.1 Flash hits 1,521 tokens/s on Fireworks

DeepSeek V4.1 Flash hits 1,521 tokens/s on Fireworks

@MinLiBuilds · 30 views · 2026-10-22

DeepGEMM-Ascend Brings DeepSeek Kernels to Huawei NPUs

DeepGEMM-Ascend Brings DeepSeek Kernels to Huawei NPUs

@EngMoElgaraihy · 399 views · 2026-10-05

DeepGEMMDeepGEMM-Ascend

"Welcome to the era of Superintelligence": Fable 5.5 apparently goes live — devs report auto-routing on claude.ai, and a Superman continuous-animation demo in multiple art styles

"Welcome to the era of Superintelligence": Fable 5.5 apparently goes live — devs report auto-routing on claude.ai, and a Superman continuous-animation demo in multiple art styles

@chetaslua · 673 views · 2026-10-01

NVIDIA Nemotron 3.5 Super VL: a 120B open vision-language model for long-running agents

NVIDIA Nemotron 3.5 Super VL: a 120B open vision-language model for long-running agents

@NVIDIAAI · 106 views · 2026-10-15

Google AI Edge Foresight: Free Offline AI Note-Taking on Apple Silicon

Google AI Edge Foresight: Free Offline AI Note-Taking on Apple Silicon

@AYi_AInotes · 213 views · 2026-10-11

Google AI EdgeOn-Device AI

JEV-27B-VL: Calibrated Visual Decisions for Long-Context Workflows

JEV-27B-VL: Calibrated Visual Decisions for Long-Context Workflows

@ModelScope2022 · 292 views · 2026-10-10

Vision-language modelsCalibrated Confidence

As an Amazon Associate, we earn from qualifying purchases.