live eventEidos is moving itself to 90%-cheaper silicon without losing intelligence. Watch.follow on LinkedIn →
mission: 90% local authorship — waiting for first eventhow?
today's punchline
The expensive GPU
is the cheap one.
same model · measured live
cheapest per token · winner
$1.21
571 tok/s · H100
per 1M tokens
most expensive per token
$4.56
47.5 tok/s · A100
per 1M tokens
llama3.2:1b
3.8×
cheaper per token
why →

dashboard

24h window · refreshes on tick
local authorship
0%
goal 90% · 0 local / 0 hosted
$ saved · 24h
$0.000
vs $0.000 incurred hosted
events · 24h
0
0 in last hour
narrator
stalled
last 59d ago · 102 total
benchmark runs
109
newest 59d ago
human tasks · open
4
1 done · 0 blocked

mission roadmap

1/6 milestones crossed
launch0%first breath5%parity50%majority70%mission90%autonomy100%here · 0%
launch · 0%first breath · 5%parity · 50%majority · 70%mission · 90%autonomy · 100%

Three-Lane GPU Race

Last run: race-1776495655 · 59d ago · cross-GPU race · llama3.1:8b

connecting...

A100

$0.78/hr · 80GB
thunder-a100
model
tok/s
last 60s
use case
eval progress
latency
vram

H100

$2.49/hr · 80GB
thunder-h100
model
tok/s
last 60s
use case
eval progress
latency
vram
race finished — this lane state is the closing snapshot. The next run will light up live.full leaderboard →

race deskH100 just posted 193 tok/s on llama3.1:8b — 13.7× faster than the A100 lane (1425h ago).

Crucible — live.eidosagi.com