DeepSeek v4.1 reportedly boosts long-context prefill throughput by ~40%
HankYeomans · x · 2026-10-11
User HankYeomans shared (unverified) performance numbers purportedly from DeepSeek v4.1:
- 16K prefill: throughput up from 2,195/2,345 to 2,934/3,264 tok/s (cold/warm, +34%/+39%), TTFT down from 7.32s to 5.48s.
- 32K prefill: throughput up from 2,499/2,612 to 3,532/3,508 tok/s (+41%/+34%), TTFT down from 12.81s to 9.06s.
Overall, long-context prefill shows 30-40% improvements in both throughput and time-to-first-token. Numbers come from an individual's testing; treat as unconfirmed.
More from Infra
- Qualcomm CEO predicts AI phone supercycle, smart glasses as top AI wearable — SuB8u · 2026-10-11
- Hugging Face launches a PyTorch profiling series: from torch.profiler to attention — ariG23498 · 2026-10-11
- Bain sees 183GW of new data center capacity by 2030, needing $5-6.5T in spending — Beth_Kindig · 2026-10-11
- TRL v1.15: new Triton kernel scales SFT/RL past 100k tokens on one GPU — _lewtun · 2026-10-11
- AWS finally adds a real spending limit, ending the fear of surprise bills for new accounts — DavidWells · 2026-10-11
- Google, Amazon, Meta, Apple drive India's renewables boom as data centre needs double to 32.4 GW by 2030 — SuB8u · 2026-10-11