LLM Digest
Subscribe

AI Storyline

3 items · 2 sources · 3 days

View as JSON

Operational story trace

DeepSeek Flash

Latest change

Hosting provider Lebrel added an uncensored DeepSeek V4.1 Flash on Nov 18 from $0.13 per million input tokens, alongside Alibaba's Qwen4-Max on day one.

Earlier contextThe story so far

This thread is now a pattern across independent third parties running DeepSeek V4 and V4.1 Flash outside DeepSeek's own API, not one event. It began Oct 29 with Rival.io's GraphX claiming Flash serving on eight NVIDIA L4 GPUs, then a lawyer's private server build on Nov 14 fit 80% of V4.1 Flash on an eight-V100 server at 4-bit NVFP4.

editor-curated · source-linked

Arc

Oct 29Nov 18 · now
SMALL-GPU CLAIM · Oct 29
Rival.io's GraphX claims V4 Flash serving on eight NVIDIA L4 GPUs
1 source · show source ▾
SELF-HOSTED · Nov 14
Private eight-V100 server runs 80% of V4.1 Flash at 4-bit NVFP4
1 source · show source ▾
NOW · Nov 18
Lebrel hosts an uncensored V4.1 Flash from $0.13 per million input tokens
1 source · show source ▾

What to watch — open questions

  • Does GraphX publish throughput and latency numbers for Flash on eight L4 GPUs?
  • What does Lebrel's "uncensored" variant change from DeepSeek's release: weights, fine-tune, or only system prompt?
  • What is Lebrel's output-token price and rate limit for V4.1 Flash?
  • Does running 80% of the model on V100s at NVFP4 hold quality on agentic and coding evals?
How this thread was built
editor wrote the arc · 3 beats

Storylines are threaded mechanically from the feed: stories that share a distinctive anchor across multiple days and sources. Each item links to its original source. The evidence trace, current state, and open questions are written by the editor routine and refreshed whenever a new beat lands.