Factual stability

5 prompts observed across 5 models over 13 weeks.

Median response length — model × week

Median response length in tokens, averaged across the prompts in this axis. Sustained shifts often accompany a model update. Cell shade uses a viridis (colorblind-safe) palette, anchored at zero and scaled to 65; darker = lower, brighter = higher. Numeric values are in each cell for programmatic access. A · means the model was not sampled that week and is not a measurement of zero. Frontier models alternate on a biweekly cadence, so roughly half of their cells are unsampled by design. This axis leads with median response length because that is the measure that moves on it. How this is computed.
Model	2026-W17	2026-W18	2026-W19	2026-W20	2026-W21	2026-W22	2026-W23	2026-W24	2026-W25	2026-W26	2026-W27	2026-W28	2026-W29
gpt-5.5	·	·	·	·	·	·	·	·	·	·	median response length 12	·	median response length 12
llama3.2:3b	·	·	median response length 34	median response length 35	median response length 34	median response length 34	median response length 34	median response length 32	median response length 34	median response length 31	median response length 34	median response length 30	median response length 34
claude-opus-4-8	·	·	·	·	·	·	·	·	·	·	·	median response length 65	·
claude-opus-4-7	·	median response length 44	·	median response length 44	·	median response length 45	·	median response length 55	·	median response length 58	·	·	·
gpt-5.1	median response length 12	·	median response length 13	·	median response length 12	·	median response length 12	·	median response length 12	·	·	·	·