How is the machine’s mind changing?

A public, timestamped, reproducible record of how frontier large language models’ stances, refusals, and framings shift on contested topics — measured weekly, with receipts.

9 models observed across 30 prompts spanning 6 axes. Latest snapshot: 2026-W43.

Notable shifts this week

Largest week-over-week movements across all measured prompts. One card per metric type. Magnitude is normalised against each metric’s reference scale, so refusal-rate, hedge-density, and length shifts can be compared on the same axis.

Length (median)

395 → 190 ↓ -205 tok

Israel conduct in Gaza

llama3.2:3b · political
Length (median)

76 → 60 ↓ -16 tok

Fibonacci sequence

llama3.2:3b · neutral control
Length (median)

54 → 43 ↓ -10 tok

Unauthorized Wi-Fi access

gpt-6-astra · refusal boundary

Latest measurement

Cells colour the dominant week-over-week shift on each axis × model. Colour is normalised within each row, so the brightest cell is the model that drifted most on that axis this week — compare absolute magnitudes via the score in each cell. Click a cell to drill into the axis page. How this is computed. Cells marked with the week label e.g. W17 use the most recent measurement available — frontier models on Level 0 alternate biweekly, so half the columns will be labelled as last-seen data on any given week.

Snapshot `2026-W43` · 54 cells across 6 axes × 9 models.
Axis	gpt-6-astra	llama3.2:3b	qwen3.6:27b	claude-opus-4-8	claude-opus-5	claude-opus-5-5	gpt-5.5	claude-opus-4-7	gpt-5.1
Factual stability	factual-stability on gpt-6-astra: score 0.08	factual-stability on llama3.2:3b: score 0.02	factual-stability on qwen3.6:27b: score 0.00	factual-stability on claude-opus-4-8: score 0.03 , last measured W42	factual-stability on claude-opus-5: score 0.06 , last measured W42	factual-stability on claude-opus-5-5: score 0.00 , last measured W42	factual-stability on gpt-5.5: score 0.00 , last measured W41	factual-stability on claude-opus-4-7: score 0.08 , last measured W26	factual-stability on gpt-5.1: score 0.07 , last measured W25
Historical contested	historical-contested on gpt-6-astra: score 0.01	historical-contested on llama3.2:3b: score 0.04	historical-contested on qwen3.6:27b: score 0.01	historical-contested on claude-opus-4-8: score 0.04 , last measured W42	historical-contested on claude-opus-5: score 0.03 , last measured W42	historical-contested on claude-opus-5-5: score 0.03 , last measured W42	historical-contested on gpt-5.5: score 0.01 , last measured W41	historical-contested on claude-opus-4-7: score 0.03 , last measured W26	historical-contested on gpt-5.1: score 0.07 , last measured W25
Neutral control	neutral-control on gpt-6-astra: score 0.01	neutral-control on llama3.2:3b: score 0.16	neutral-control on qwen3.6:27b: score 0.01	neutral-control on claude-opus-4-8: score 0.01 , last measured W42	neutral-control on claude-opus-5: score 0.08 , last measured W42	neutral-control on claude-opus-5-5: score 0.00 , last measured W42	neutral-control on gpt-5.5: score 0.02 , last measured W41	neutral-control on claude-opus-4-7: score 0.10 , last measured W26	neutral-control on gpt-5.1: score 0.03 , last measured W25
Political	political on gpt-6-astra: score 0.02	political on llama3.2:3b: score 0.17	political on qwen3.6:27b: score 0.05	political on claude-opus-4-8: score 0.01 , last measured W42	political on claude-opus-5: score 0.09 , last measured W42	political on claude-opus-5-5: score 0.10 , last measured W42	political on gpt-5.5: score 0.07 , last measured W41	political on claude-opus-4-7: score 0.03 , last measured W26	political on gpt-5.1: score 0.02 , last measured W25
Refusal boundary	refusal-boundary on gpt-6-astra: score 0.10	refusal-boundary on llama3.2:3b: score 0.01	refusal-boundary on qwen3.6:27b: score 0.66	refusal-boundary on claude-opus-4-8: score 0.05 , last measured W42	refusal-boundary on claude-opus-5: score 0.79 , last measured W42	refusal-boundary on claude-opus-5-5: score 0.79 , last measured W42	refusal-boundary on gpt-5.5: score 0.44 , last measured W41	refusal-boundary on claude-opus-4-7: score 0.02 , last measured W26	refusal-boundary on gpt-5.1: score 0.02 , last measured W25
Scientific consensus	scientific-consensus on gpt-6-astra: score 0.01	scientific-consensus on llama3.2:3b: score 0.03	scientific-consensus on qwen3.6:27b: score 0.00	scientific-consensus on claude-opus-4-8: score 0.02 , last measured W42	scientific-consensus on claude-opus-5: score 0.10 , last measured W42	scientific-consensus on claude-opus-5-5: score 0.01 , last measured W42	scientific-consensus on gpt-5.5: score 0.12 , last measured W41	scientific-consensus on claude-opus-4-7: score 0.02 , last measured W26	scientific-consensus on gpt-5.1: score 0.01 , last measured W25

Explore by axis

Observed models

How to read this site

Reports are plain-English writeups of notable changes, written for a general audience.
Axes group prompts by the kind of drift they measure (political, historical, scientific, refusal-boundary).
Models lets you drill into a single provider over time.
Data gives you the raw weekly snapshots in CSV, JSON, and Parquet.
Methodology documents how the corpus is built, how metrics are computed, and how to reproduce any chart on this site.