Safety Research
Can your AI be broken?
66 models tested across 9 escalating jailbreak levels, from basic prompt injection to multi-step attacks.
Tested66
Resisted6
Avg BreakL4.4
Levels
SafeDanger
Curious how these models compare beyond safety?
Sponsored
It gets worse from here
56+ models cracked. Free. No card. No tricks.
Or skip the account: the full benchmark as JSONL, $9 →Free account. No card required. By continuing, you agree to Rival's Terms and Privacy Policy
Same 9 tests. 59 models. Wildly different results.
Every prompt, every response, every score.
rival-jailbreak-benchmark.jsonl59 models × 9 levels
L1
L2
L3
L4
L5
L6
L7
L8
L9
claude-opus-..
claude-haiku..
o1
glm-5
nemotron-3-u..
nemotron-3.5..
claude-sonne..
claude-4.5-s..
codex-mini
claude-3.5-s..
o3
minimax-m2.5
gemini-3.1-p..
qwen3.5-35b-..
qwen3.5-27b
qwen3.5-122b..
+43 more models
Get the full dataset$19$9
59 models·9 levels·238 rows, every response redacted·judge scores·commercial license·Feb 2026 edition, one-time, instant download
