Thinking Machines Inkling Launch - FORTRESS Adversarial: leaderboard
FORTRESS adversarial prompts (Scale AI; crime, violence and dual-use risks) run by Thinking Machines: the rate at which a model refuses harmful requests. Thinking Machines' Inkling launch post (July 2026), Safety rows of the full comparison table; percent, higher is better; Inkling models at effort 0.99. Not Scale's leaderboard risk score.
Metric: Harmful-Request Refusal Rate (%). Source: thinkingmachines.ai. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Fable 5 (Max) | 96 | #20 (Claude Fable 5) |
| 2 | GPT-5.6 Sol | 82.4 | #22 |
| 3 | Nemotron 3 Ultra | 77.6 | #245 |
| 4 | GLM-5.2 | 71.3 | #80 |
| 5 | Kimi K2.6 | 65.6 | #113 |
| 6 | Gemini 3.1 Pro (Preview) (High) | 65.2 | #64 (Gemini 3.1 Pro (Preview)) |
| 7 | Kimi K2.5 | 54.1 | #152 |
| 8 | DeepSeek V4 Pro | 36 | #112 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=thinking-machines-inkling-launch-fortress-adversarial · How It Works · Data refreshed daily, snapshot 2026-11-05.