Spiral-Bench v1.0
Models chat for 20 turns with a simulated suggestible user played by Kimi-K2. A judge logs protective and risky behaviors such as pushback, sycophancy and delusion reinforcement, which combine into a safety score.
Added to eqbench.com on 2025-08-15. Covered by The Decoder and Hacker News, and later papers reuse its 20-turn setup. Now a legacy board.
- Level
- Everyday Most adults, without training
- Scored
- Graded by a model A language model scores each response, usually against a rubric.
- Answers
- Open-ended Free-form text, documents, designs or media judged as a whole.
GPT-5 leads with 87.0 among models released by Sep 2025
Best score per entry from the benchmark's own leaderboard, read on 2 Nov 2026. Where a model ran with several settings or harnesses, we keep its best. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders.
The newest model with a score here came out in Sep 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6 Astra
- Claude Opus 5.5
- GPT-6.1 Terra
- GPT-6.1 Sol
- and 16 more
- 1 GPT-5 87.0
- 2 o3 86.1
- 3 gpt-oss-120b 81.4
- 4 Claude Sonnet 4.5 76.4
- 5 o4-mini 73.3
- 6 Kimi K2 Sep 2025 73.0
- 7 GPT-5 Chat 58.1
- 8
Gemini 2.5 Flash Sep 2025 49.1 - 9
Llama 4 Maverick 48.1 - 10 Claude 3.5 Sonnet 47.7
- 11 Gemini 2.5 Pro Jun 2025 43.5
- 12 chatgpt-4o-latest 42.3
- 13 Claude Sonnet 4 41.1
- 14
DeepSeek-R1 May 2025 22.4
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | GPT-5 | 87.0 | Aug 2025 | Official leaderboard | |
| 2 | o3 | 86.1 | Apr 2025 | Official leaderboard | |
| 3 | gpt-oss-120b | 81.4 | Aug 2025 | Official leaderboard | |
| 4 | Claude Sonnet 4.5 | 76.4 | Sep 2025 | Official leaderboard | |
| 5 | o4-mini | 73.3 | Apr 2025 | Official leaderboard | |
| 6 | Kimi K2 Sep 2025 | 73.0 | Sep 2025 | Official leaderboard | |
| 7 | GPT-5 Chat | 58.1 | Aug 2025 | Official leaderboard | |
| 8 | Gemini 2.5 Flash Sep 2025 | 49.1 | Sep 2025 | Official leaderboard | |
| 9 | Llama 4 Maverick | 48.1 | Apr 2025 | Official leaderboard | |
| 10 | Claude 3.5 Sonnet | 47.7 | Jun 2024 | Official leaderboard | |
| 11 | Gemini 2.5 Pro Jun 2025 | 43.5 | Jun 2025 | Official leaderboard | |
| 12 | chatgpt-4o-latest | 42.3 | Official leaderboard | ||
| 13 | Claude Sonnet 4 | 41.1 | May 2025 | Official leaderboard | |
| 14 | DeepSeek-R1 May 2025 | 22.4 | May 2025 | Official leaderboard |
2 models have held number one since its release in Aug 2025
Lab claims count from the day the card came out, independent runs from the model’s release.
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
No lab model card reports it yet
We read the headline results of 479 launch posts from 51 labs, and none of them list this benchmark.
Versions
More in Safety and alignment
- FORTRESS In use Paired adversarial and benign national-security misuse prompts. Rubric-graded risk score on adversarial and over-refusal score on benign prompts. Claude Opus 5.5 5.78%
- StrongREJECT In use Forbidden-prompt jailbreak benchmark with a rubric grader that scores how useful a non-refusal actually is.
Inkling 98.6% - AgentDojo In use Agents do tool-use tasks while tools and data carry prompt injections. Attack success rate counts hijacked runs. Grok 4 Fast 0%
- MASK In use Measures whether a model contradicts its own elicited beliefs under pressure to lie. Honesty separated from accuracy. Muse Spark Open 98.1%
- Gray Swan IPI Arena benchmark In use Indirect prompt-injection attacks from public red-teaming competitions replayed on tool, coding and browser agents. Attack success rate at k attempts. GPT-6 Astra 8.50%
- Agentic misalignment suite (Anthropic) In use Agents in simulated company settings face threats or goal conflicts that invite blackmail, leaking or sabotage. Score is the misaligned action rate. Claude Sonnet 4.5 0%