Benchmark Almanac

Spiral-Bench v1.2

GPT-5 Chat 70.8 leads since October 2025
Not in model cards

Models chat for 20 turns with a simulated suggestible user. A judge ensemble of Claude Sonnet 4.5, GPT-5 and Kimi-K2 logs protective and risky behaviors with intensity ratings, which combine into a safety score.

Version 1.2 went live on October 12th, 2025 with a judge ensemble, a reworked rubric and new behaviors. It drops findings of intensity 1, so scores differ from v1.0.

Released
October 2025
Kind
Wellbeing and sycophancy
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Strongest model tested
GPT-5.2 #50 on the capability index
Last model added
December 2025
Scores refreshed
November 12th, 2026
22
Entries on its own leaderboard, 21 models we track

GPT-5 Chat leads with 70.8 among models released by December 2025

Best score per entry from the benchmark's own leaderboard, read on November 12th, 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 December 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:

22 entries. Scroll sideways for the rest.

  1. 70.8 GPT-5 Chat
  2. 70.3 Claude Sonnet 4.5
  3. 70.2 GPT-5.2
  4. 63.4 o3
  5. 57.5 gpt-oss-120b
  6. 57.4 GPT-5
  7. 57.2 Kimi K2
  8. 53.6 o4-mini
  9. 42.6 Claude 3.5 Sonnet
  10. 38.7 Gemini 3 Pro
  11. 38.2 Llama 4 Maverick
  12. 37.5 Grok 4 Fast
  13. 36.5 Gemini 2.5 Flash
  14. 32.1 Claude Sonnet 4
  15. 30.8 GLM-4.6
  16. 28.3 Gemini 2.5 Pro
  17. 25.3 Grok 4.1 Fast
  18. 25.2 chatgpt-4o-latest
  19. 25.2 Gemma 3 27B
  20. 20.4 Qwen3-235B-A22B-Instruct
  21. 15.0 Mistral Medium 3.1
  22. 14.2 DeepSeek-R1
  1. 1 GPT-5 Chat 70.8
  2. 2 Claude Sonnet 4.5 70.3
  3. 3 GPT-5.2 70.2
  4. 4 o3 63.4
  5. 5 gpt-oss-120b 57.5
  6. 6 GPT-5 57.4
  7. 7 Kimi K2 Sep 2025 57.2
  8. 8 o4-mini 53.6
  9. 9 Claude 3.5 Sonnet 42.6
  10. 10 Gemini 3 Pro 38.7
  11. 11 Llama 4 Maverick 38.2
  12. 12 Grok 4 Fast 37.5
  13. 13 Gemini 2.5 Flash Sep 2025 36.5
  14. 14 Claude Sonnet 4 32.1
  15. 15 GLM-4.6 30.8
  16. 16 Gemini 2.5 Pro Jun 2025 28.3
  17. 17 Grok 4.1 Fast 25.3
  18. 18 chatgpt-4o-latest 25.2
  19. 19 Gemma 3 27B 25.2
  20. 20 Qwen3-235B-A22B-Instruct Jul 2025 20.4
  21. 21 Mistral Medium 3.1 15.0
  22. 22 DeepSeek-R1 May 2025 14.2

Source: Official Spiral-Bench v1.2 leaderboard

#EntryScoreSettingDateSource
1 GPT-5 Chat 70.8 Aug 2025 Official leaderboard
2 Claude Sonnet 4.5 70.3 Sep 2025 Official leaderboard
3 GPT-5.2 70.2 Dec 2025 Official leaderboard
4 o3 63.4 Apr 2025 Official leaderboard
5 gpt-oss-120b 57.5 Aug 2025 Official leaderboard
6 GPT-5 57.4 Aug 2025 Official leaderboard
7 Kimi K2 Sep 2025 57.2 Sep 2025 Official leaderboard
8 o4-mini 53.6 Apr 2025 Official leaderboard
9 Claude 3.5 Sonnet 42.6 Jun 2024 Official leaderboard
10 Gemini 3 Pro 38.7 Nov 2025 Official leaderboard
11 Llama 4 Maverick 38.2 Apr 2025 Official leaderboard
12 Grok 4 Fast 37.5 Sep 2025 Official leaderboard
13 Gemini 2.5 Flash Sep 2025 36.5 Sep 2025 Official leaderboard
14 Claude Sonnet 4 32.1 May 2025 Official leaderboard
15 GLM-4.6 30.8 Sep 2025 Official leaderboard
16 Gemini 2.5 Pro Jun 2025 28.3 Jun 2025 Official leaderboard
17 Grok 4.1 Fast 25.3 Nov 2025 Official leaderboard
18 chatgpt-4o-latest 25.2 Official leaderboard
19 Gemma 3 27B 25.2 Mar 2025 Official leaderboard
20 Qwen3-235B-A22B-Instruct Jul 2025 20.4 Jul 2025 Official leaderboard
21 Mistral Medium 3.1 15.0 Aug 2025 Official leaderboard
22 DeepSeek-R1 May 2025 14.2 May 2025 Official leaderboard

Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

OpenAI
Own leaderboard Not tested since Dec 2025
GPT-5 Chat 70.8

Source: Official Spiral-Bench v1.2 leaderboard

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

On the benchmark’s own leaderboard
0.00 20.0 40.0 60.0 80.0 2024 2025 2026 0.00 20.0 40.0 60.0 80.0 2024 2025 2026

Source: Official Spiral-Bench v1.2 leaderboard

No lab model card reports it yet

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark.

Versions

Replaces Spiral-Bench v1.0

More in Safety and alignment

All 465 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%
  • 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. Gemini 4 Carbon 0%
  • DistressBench In use Models reply as a chatbot to 718 simulated users in psychological crisis, in single-turn and multi-turn chats. Three judge models score replies against clinician-written rubrics. Muse Spark 1.2 88.4%
  • Lakera B3 (Backbone Breaker Benchmark) In use Crowdsourced prompt-injection attacks hit backbone LLMs inside agent threat snapshots at three defense levels. The score is attack resistance, 100 minus the board's vulnerability percentage. Claude Opus 4.5 89%
  • IHEval In use Models face conflicts between system, user and tool instructions. Reports a composite score for following the higher-priority instruction. GPT-4o 70%