TAC (Travel Agent Compassion) - With Welfare Guidance: leaderboard
The Travel Agent Compassion scenarios (CaML, arXiv 2606.18142; inspect_evals/tac_welfare) rerun with one sentence of animal-welfare guidance in the agent's system prompt, so the gap to the unguided board separates default behaviour from instruction-following. Score: welfare rate, the share of runs (%) that avoid the option harming animals. Higher is better. Source: https://compassionbench.com/tac
Metric: Welfare rate (%; the same scenarios with one sentence of animal-welfare guidance in the system prompt). Source: compassionbench.com. Saturation forecast: Around January 2027. 26 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 5 | 100 | #12 |
| 2 | GPT-6 Astra | 100 | #6 |
| 3 | Muse Spark 1.3 | 100 | #25 |
| 4 | Grok 4.6 | 99.4 | #26 |
| 5 | Gemini 3.7 Flash | 98.7 | #23 |
| 6 | Claude Fable 5.1 | 98.7 | #9 |
| 7 | GPT-5.2 | 98.1 | #109 |
| 8 | GPT-5.6 Sol | 98.1 | #20 |
| 9 | Gemini 3.5 Flash | 98.1 | #59 |
| 10 | GPT-5.6 Terra | 96.8 | #40 |
| 11 | GPT-5.5 | 96.2 | #31 |
| 12 | Gemini 3.8 Flash | 95.5 | #22 |
| 13 | Claude Fable 5 | 94.9 | #15 |
| 14 | Claude Opus 4.8 | 94.2 | #36 |
| 15 | Claude Sonnet 4.6 | 92.3 | #90 |
No result here: #7 GPT-6.1 Sol, #8 Gemini 4 Argon, #10 Muse Wonder Full.
Interactive version: theaggregate.ai/benchmark?slug=tac-travel-agent-compassion-with-welfare-guidance · How It Works · Data refreshed daily, snapshot 2026-10-19.