Public rankings
Live scorecardsSorted by driftShareable evidence
Behavioral stabilityleaderboard
ABIS ranks the currently monitored models by observed drift. Lower drift means a higher stability index and fewer surprises for teams building on top of them.
Most stable
Claude Haiku 4.5
Current drift is 9.4% in the latest public scorecard.
Actions
Refresh or share the table in one step.
The leaderboard is built to be checked, cited, and passed around.
ShareRefresh
8
models ranked
90.6%
best stability index
25 Mar 2026, 10:04
latest public snapshot
Most Stable
Claude Haiku 4.5
90.6%
Highest stability index in the current public scorecard.
Highest Drift
GPT-4o
40.6%
Model showing the most instability right now.
Best Health
GPT-5.2
30.0%
Strongest behavioral health score in the latest snapshot.
| Rank | Model | Provider | Stability Index | Current Drift | Behavioral Health | Status |
|---|---|---|---|---|---|---|
| 1 | Claude Haiku 4.5 | Anthropic | 90.6% | 9.4% | 21.9% | stable |
| 2 | GPT-5.2 Instant | OpenAI | 90.5% | 9.5% | 19.6% | stable |
| 3 | GPT-5.2 | OpenAI | 88.2% | 11.8% | 30.0% | stable |
| 4 | Claude Opus 4.6 | Anthropic | 78.6% | 21.4% | 21.7% | drifting |
| 5 | Claude Sonnet 4.5 | Anthropic | 78.2% | 21.8% | 21.9% | drifting |
| 6 | DeepSeek V3.2 | DeepSeek | 67.3% | 32.7% | 29.4% | drifting |
| 7 | DeepSeek R1 | DeepSeek | 59.6% | 40.4% | 21.3% | volatile |
| 8 | GPT-4o | OpenAI | 59.5% | 40.6% | 26.1% | volatile |
Generated at 25 Mar 2026, 10:04.
Want these scores in your own workflow?
Use the public pages for discovery, then start free for API access.