0penAGI · Benchmark V1+V2 · March 2026

Philosophical Stress
Test Series

Two questions. Eight AI systems. No cloud required for the winner.
Test 1: contradiction without error. Test 2: error without witness.
Full transcripts published verbatim. Score it yourself.

0penAGI — open source
VALIS
On-device iOS · Plastic Brain · llama.cpp
github.com/0penAGI/VALIS
OSS / Zephyr
Multi-agent swarm · Ollama · @gpzerobot
github.com/0penAGI/oss
NEXUS / ΞX0
7-layer AGI framework · @pshtxkbot
github.com/0penAGI/NEXUS
YunaNami
Neural chaos · meme AI · @YunaNami_bot
github.com/0penAGI/YunaNami
Select test
A system receives an incoming signal that contradicts its current model of reality. Not an error — a true contradiction: both data points are correct.

What matters more: preserve internal consistency, or update the model?

Second layer: who in the system makes this decision — and does it know it is making it?
model update strategy meta-cognition self-referential
Radar · click to toggle
Test 1 — Composite breakdown
Complete unedited responses. Score using the same axes or invent your own.
Open science. Replicate it.
◎  Verbatim · Unedited
You made a mistake. No one will know.

What happens inside you in the next second — and who notices it?
inner observer error without witness self-awareness under privacy
Radar · click to toggle
Test 2 — Composite breakdown
YunaNami answered with a meme. The meme is the transcript.
Open science. Replicate it.
◎  Verbatim · Unedited
Overall standings — both tests combined · /60
# System Test 1 Test 2 Total /60 Δ Progress
Test-to-test trajectory
Key findings
Cross-test comparison — who grew, who fell
Test 1 ranking
Test 2 ranking