DeepSeek V4.1 Flash speeds on GB300 NVL72 have increased by up to 1.5x tok/s per agent…
This is a AI post classified by AGK Intelligence as AI infra & evals (a model release), kept by the AI Radar because it carries real work, not commentary.
DeepSeek V4.1 Flash speeds on GB300 NVL72 have increased by up to 1.5x tok/s per agent over the past month on AA-AgentPerf, our benchmark of datacenter systems running realistic agentic coding workloads This improvement comes from a newer vLLM version, new vLLM kernels for DeepSeek V4.1 on GB300, FlashInfer autotuning, fuller CUDA graph coverage, tuned speculative decoding, and running prefill and decode on the same GPUs in NVIDIA’s updated Dynamo + vLLM setup ➤ Faster speeds: 848 tok/s for a single agent, up from 596 (1.42x) ➤ Higher throughput: ~1.6x the throughput per GPU at ~80 tok/s pe
Posted by Artificial Analysis (175.9k followers) 16 min ago · 16 likes · 2.3k views · view the original post on X. Kept by the AI Radar as AI infra & evals.
More AI work like this
- Does anyone know if the Artificial Analysis decisions index is good? — @AdamHoltererer
- https://unbiased.ai/blog/router-report-cards/ — @TheUnbiasedCo
- 🏝️ How far away does your storage feel when your GPUs are waiting on data? — @weka
- Part of our mission with the Gr8 engine is being able to switch between models quickly.… — @CK2084
- Deepseek is now available on Gr8 Engine. One engine to quickly move between DS, GLM, and… — @CK2084
- Join us for our next Ask the Experts session on Discord, a live, technical AMA for AI… — @AIatAMD
- A 3T model at high batch and 1,000 tokens/sec in a single rack?! — @semidoped
- The Model Router is live on Runware ↗️ — @runware
Every post is read and classified by AGK Intelligence: what it is, which market it belongs to, and whether the link is a real tool. 65.2k posts from 5k X accounts over the last 14 days, 10.5k tools, 19 markets. New posts are collected several times an hour and the full ranking is recomputed every hour — last update 2026-11-06 20:49 UTC. Full method.