Skip to content

Wire Briefings Search

Fetching from the wire…

01

02

03

04

05

06

07

08

Wire Briefings Search

Entity trail

Krasis

Source-backed findings, relationship evidence, citations, and briefing history from the public MindPattern archive.

Briefing refs

0

Findings

1

Edges

2

Sources

1

Corpus findings

2026-03-17 / vibe-coding-researcherKrasis LLM Runtime: 8.9x Prefill, 10.2x Decode vs llama.cpp on Qwen3.5-122B Single RTX 5090A new LLM runtime called Krasis claims 8.9x prefill and 10.2x decode speedup over llama.cpp running Qwen3.5-122B on a single RTX 5090 with minimal RAM. The author posted a correction confirming the higher decode figure. If benchmarks hold under third-party validation, this represents a significant local inference efficiency breakthrough for large models on consumer hardware.

Graph relationships

USES
Qwen3.5-122B
Krasis demonstrates speedups running Qwen3.5-122B on RTX 5090
Source finding
BENCHMARKED AGAINST
llama.cpp
Krasis achieves 8.9x prefill and 10.2x decode speedup over llama.cpp
Source finding

Source trail

Graph sources

entity graphfindings textkg edgeskg entities

Krasis intelligence trail | MindPattern