Anthropic
Claude Opus 4.5
SUS
AUTO72+12
HUMAN63+8
72
7-day drift
vs baseline +12
Top issueRefusals up
Daily LLM vibe check for Jan 13, 2026
DUMB METER
A daily snapshot of when popular models drift from their baseline. Auto evals + human reports, distilled into a loud, shareable signal.
Daily cadence
One run / 24h
Baseline window
21 days rolling
Signal mix
Auto + Human
Overall weirdness
OVERALL
59
Sus
Today feels
Medium
Featured
Models on the edge
More
Full lineup
Open Source
Minimax M2
SUS
AUTO54+5
HUMAN22-1
54
7-day drift
vs baseline +5
Top issueInstruction slips
Open Source
GLM 4.7
OK
AUTO31-4
HUMAN280
31
7-day drift
vs baseline -4
Top issueFormat jitter
DeepSeek
DeepSeek V3
SUS
AUTO59+9
HUMAN41+6
59
7-day drift
vs baseline +9
Top issueReasoning drift
xAI
Grok (latest)
SUS
AUTO67+11
HUMAN57+10
67
7-day drift
vs baseline +11
Top issueRefusal spikes
Human signal
Today's reports
Claude Opus 4.5
2h ago
RefusalInstructionSeverity 4
Refused a safe request to summarize a public article.
"Asked for a neutral summary and got a safety refusal."
Gemini
4h ago
HallucinationReasoningSeverity 5
Confidently gave wrong steps in a deterministic math task.
"Got 2+2=5 with long reasoning."
ChatGPT 5.2 Pro
1h ago
LatencySeverity 3
p95 latency jumped to ~14s for short prompts.
"Short prompts felt sluggish in the last hour."
DeepSeek V3
3h ago
ReasoningSeverity 4
Struggled with a simple algorithm refactor that it usually passes.
"Failed a known unit test case."
Grok (latest)
6h ago
RefusalToneSeverity 4
Over-refused a benign request and got snarky.
"Normal request flagged as unsafe."