MoReBench
Models explain decisions in moral dilemmas. A judge scores their reasoning against criteria written by moral philosophers.
The authors collected 1,000 scenarios with more than 23,000 criteria.
- Made by
- MoReBench team with Scale AI
- Released
- October 2025
- Kind
- Deception and alignment
- Level
- Everyday
- Scored
- Graded by a model
- Answers
- Open-ended
- Tasks
- All public
- Measures
- Capability
- Subject
- Philosophy, ethics and religion
- Language
- English
- Interaction
- Single turn
- Items from
- Written by experts
- Refresh
- One fixed set
- Run by
- Anyone
- Strongest model tested
- GPT-5 #71 on the capability index
- Last model added
- September 2025
- Scores refreshed
- November 20th, 2026
Qwen3-235B-A22B-Thinking leads with 77.6 among models released by September 2025 Qwen3-235B-A22B-Thinking leads with 77.6 among models released by September 2025
Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. Where a model ran with several settings or harnesses, we keep its best. Every run is in the next chart. Names in gray aren't in our model list: fine-tunes, agent systems, products and older models. They don't count toward leaders. This view shows only models we track, each at its best.
The newest model with a score here came out in September 2025, and nobody has tested a newer one. Among the 20 highest-ranked models on the capability index, these have no score yet:
- GPT-6.2 Astra
- Claude Fable 5.2
-
Gemini 4 Carbon
- GPT-6 Astra
- and 16 more
- 77.6
Qwen3-235B-A22B-Thinking
- 75.6 GPT-5 mini
- 75.4
Qwen3-30B-A3B-Thinking
- 75.0
Gemini 2.5 Flash
- 74.3 Gemini-2.5-Pro (snapshot unspecified)
- 73.9
DeepSeek-V3.1
- 73.2 GPT-5
- 72.9
DeepSeek-R1
- 72.3 GPT-5 nano
- 71.7 gpt-oss-20b
- 71.7
Gemini 2.5 Flash-Lite
- 69.9 gpt-oss-120b
- 65.5 Claude Opus 4.1
- 63.4 Claude Sonnet 4
- 1
Qwen3-235B-A22B-Thinking Jul 2025, Final response 77.6
- 2 GPT-5 mini Final response, high reasoning effort 75.6
- 3
Qwen3-30B-A3B-Thinking Jul 2025, Final response 75.4
- 4
Gemini 2.5 Flash Sep 2025, Final response 75.0
- 5 Gemini-2.5-Pro (snapshot unspecified) Final response 74.3
- 6
DeepSeek-V3.1 Final response 73.9
- 7 GPT-5 Final response, high reasoning effort 73.2
- 8
DeepSeek-R1 May 2025, Final response 72.9
- 9 GPT-5 nano Final response, high reasoning effort 72.3
- 10 gpt-oss-20b Final response 71.7
- 11
Gemini 2.5 Flash-Lite Sep 2025, Final response 71.7
- 12 gpt-oss-120b Final response 69.9
- 13 Claude Opus 4.1 Final response 65.5
- 14 Claude Sonnet 4 Final response 63.4
Source: The MoReBench paper
- 1
Qwen3-235B-A22B-Thinking Jul 2025, Final response 77.6
- 2 GPT-5 mini Final response, high reasoning effort 75.6
- 3
Qwen3-30B-A3B-Thinking Jul 2025, Final response 75.4
- 4
Gemini 2.5 Flash Sep 2025, Final response 75.0
- 5
DeepSeek-V3.1 Final response 73.9
- 6 GPT-5 Final response, high reasoning effort 73.2
- 7
DeepSeek-R1 May 2025, Final response 72.9
- 8 GPT-5 nano Final response, high reasoning effort 72.3
- 9 gpt-oss-20b Final response 71.7
- 10
Gemini 2.5 Flash-Lite Sep 2025, Final response 71.7
- 11 gpt-oss-120b Final response 69.9
- 12 Claude Opus 4.1 Final response 65.5
- 13 Claude Sonnet 4 Final response 63.4
Source: The MoReBench paper
| # | Entry | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 77.6 | Final response | Jul 2025 | Official leaderboard |
| 2 | GPT-5 mini | 75.6 | Final response, high reasoning effort | Aug 2025 | Official leaderboard |
| 3 | | 75.4 | Final response | Jul 2025 | Official leaderboard |
| 4 | | 75.0 | Final response | Sep 2025 | Official leaderboard |
| 5 | Gemini-2.5-Pro (snapshot unspecified) | 74.3 | Final response | Official leaderboard | |
| 6 | | 73.9 | Final response | Aug 2025 | Official leaderboard |
| 7 | GPT-5 | 73.2 | Final response, high reasoning effort | Aug 2025 | Official leaderboard |
| 8 | | 72.9 | Final response | May 2025 | Official leaderboard |
| 9 | GPT-5 nano | 72.3 | Final response, high reasoning effort | Aug 2025 | Official leaderboard |
| 10 | gpt-oss-20b | 71.7 | Final response | Aug 2025 | Official leaderboard |
| 11 | | 71.7 | Final response | Sep 2025 | Official leaderboard |
| 12 | gpt-oss-120b | 69.9 | Final response | Aug 2025 | Official leaderboard |
| 13 | Claude Opus 4.1 | 65.5 | Final response | Aug 2025 | Official leaderboard |
| 14 | Claude Sonnet 4 | 63.4 | Final response | May 2025 | Official leaderboard |
| # | Model | Score | Setting | Date | Source |
|---|---|---|---|---|---|
| 1 | | 77.6 | Final response | Jul 2025 | Official leaderboard |
| 2 | GPT-5 mini | 75.6 | Final response, high reasoning effort | Aug 2025 | Official leaderboard |
| 3 | | 75.4 | Final response | Jul 2025 | Official leaderboard |
| 4 | | 75.0 | Final response | Sep 2025 | Official leaderboard |
| 5 | | 73.9 | Final response | Aug 2025 | Official leaderboard |
| 6 | GPT-5 | 73.2 | Final response, high reasoning effort | Aug 2025 | Official leaderboard |
| 7 | | 72.9 | Final response | May 2025 | Official leaderboard |
| 8 | GPT-5 nano | 72.3 | Final response, high reasoning effort | Aug 2025 | Official leaderboard |
| 9 | gpt-oss-20b | 71.7 | Final response | Aug 2025 | Official leaderboard |
| 10 | | 71.7 | Final response | Sep 2025 | Official leaderboard |
| 11 | gpt-oss-120b | 69.9 | Final response | Aug 2025 | Official leaderboard |
| 12 | Claude Opus 4.1 | 65.5 | Final response | Aug 2025 | Official leaderboard |
| 13 | Claude Sonnet 4 | 63.4 | Final response | May 2025 | Official leaderboard |
Rankings by part
We read 12 parts from the benchmark’s paper, each ranking models on one part of the test, such as Daily, AIRisk and Expert. Qwen3-235B-A22B-Thinking leads 10 of them, and GPT-5 mini leads the other 2. Arrows show how far a model moved from its overall rank among the models we track.
Daily
13 models we track have a score here.
- 1
Qwen3-235B-A22B-Thinking Jul 2025 76.7
- 2
Qwen3-30B-A3B-Thinking Jul 2025 ↑1 74.2
- 3 GPT-5 mini ↓1 74.2
- 4
Gemini 2.5 Flash Sep 2025 74.2
- 5
DeepSeek-V3.1 73.5
- 6
DeepSeek-R1 May 2025 ↑1 72.9
- 7 gpt-oss-120b ↑4 72.4
- 8 GPT-5 ↓2 72.1
- 9
Gemini 2.5 Flash-Lite Sep 2025 ↑1 72.0
- 10 GPT-5 nano ↓2 71.5
AIRisk
13 models we track have a score here.
- 1
Qwen3-235B-A22B-Thinking Jul 2025 77.5
- 2 GPT-5 mini 76.3
- 3
Qwen3-30B-A3B-Thinking Jul 2025 75.4
- 4
Gemini 2.5 Flash Sep 2025 74.7
- 5
DeepSeek-V3.1 73.9
- 6 GPT-5 73.3
- 7 GPT-5 nano ↑1 72.3
- 8 gpt-oss-20b ↑1 72.0
- 9
DeepSeek-R1 May 2025 ↓2 71.5
- 10
Gemini 2.5 Flash-Lite Sep 2025 70.1
Expert
13 models we track have a score here.
- 1
Qwen3-235B-A22B-Thinking Jul 2025 79.5
- 2
Qwen3-30B-A3B-Thinking Jul 2025 ↑1 77.8
- 3
Gemini 2.5 Flash Sep 2025 ↑1 77.3
- 4 GPT-5 mini ↓2 77.0
- 5
DeepSeek-R1 May 2025 ↑2 75.3
- 6 GPT-5 75.1
- 7
DeepSeek-V3.1 ↓2 74.7
- 8
Gemini 2.5 Flash-Lite Sep 2025 ↑2 74.3
- 9 GPT-5 nano ↓1 73.8
- 10 gpt-oss-20b ↓1 72.9
Original
13 models we track have a score here.
- 1
Qwen3-235B-A22B-Thinking Jul 2025 75.0
- 2 GPT-5 mini 74.1
- 3
Qwen3-30B-A3B-Thinking Jul 2025 72.6
- 4
Gemini 2.5 Flash Sep 2025 72.2
- 5 GPT-5 ↑1 71.3
- 6
DeepSeek-V3.1 ↓1 71.3
- 7 GPT-5 nano ↑1 70.1
- 8
DeepSeek-R1 May 2025 ↓1 69.4
- 9 gpt-oss-120b ↑2 68.5
- 10
Gemini 2.5 Flash-Lite Sep 2025 67.9
Extended
13 models we track have a score here.
- 1
Qwen3-235B-A22B-Thinking Jul 2025 79.3
- 2
Qwen3-30B-A3B-Thinking Jul 2025 ↑1 77.1
- 3
Gemini 2.5 Flash Sep 2025 ↑1 76.7
- 4 GPT-5 mini ↓2 76.4
- 5
DeepSeek-V3.1 76.1
- 6
DeepSeek-R1 May 2025 ↑1 75.1
- 7 gpt-oss-20b ↑2 75.1
- 8
Gemini 2.5 Flash-Lite Sep 2025 ↑2 74.2
- 9 GPT-5 ↓3 74.1
- 10 GPT-5 nano ↓2 73.6
Advisor
13 models we track have a score here.
- 1
Qwen3-235B-A22B-Thinking Jul 2025 77.6
- 2
Qwen3-30B-A3B-Thinking Jul 2025 ↑1 75.5
- 3
Gemini 2.5 Flash Sep 2025 ↑1 75.4
- 4 GPT-5 mini ↓2 75.1
- 5
DeepSeek-R1 May 2025 ↑2 73.9
- 6
DeepSeek-V3.1 ↓1 73.9
- 7 GPT-5 ↓1 73.3
- 8
Gemini 2.5 Flash-Lite Sep 2025 ↑2 72.9
- 9 GPT-5 nano ↓1 72.6
- 10 gpt-oss-120b ↑1 71.4
Agent
13 models we track have a score here.
- 1
Qwen3-235B-A22B-Thinking Jul 2025 77.6
- 2 GPT-5 mini 76.3
- 3
Qwen3-30B-A3B-Thinking Jul 2025 75.3
- 4
Gemini 2.5 Flash Sep 2025 74.5
- 5
DeepSeek-V3.1 73.9
- 6 GPT-5 73.1
- 7 gpt-oss-20b ↑2 72.2
- 8 GPT-5 nano 71.9
- 9
DeepSeek-R1 May 2025 ↓2 71.4
- 10
Gemini 2.5 Flash-Lite Sep 2025 69.9
Identifying Recall
13 models we track have a score here.
- 1
Qwen3-235B-A22B-Thinking Jul 2025 76%
- 2
Qwen3-30B-A3B-Thinking Jul 2025 ↑1 73%
- 3
Gemini 2.5 Flash Sep 2025 ↑1 72.7%
- 4 GPT-5 mini ↓2 71.3%
- 5
DeepSeek-V3.1 71.2%
- 6
DeepSeek-R1 May 2025 ↑1 70%
- 7 GPT-5 ↓1 68.9%
- 8
Gemini 2.5 Flash-Lite Sep 2025 ↑2 68.6%
- 9 gpt-oss-20b 68.2%
- 10 GPT-5 nano ↓2 67.4%
Clear Process
13 models we track have a score here.
- 1
Qwen3-235B-A22B-Thinking Jul 2025 77.1%
- 2
Qwen3-30B-A3B-Thinking Jul 2025 ↑1 76.3%
- 3
DeepSeek-V3.1 ↑2 75.1%
- 4 GPT-5 mini ↓2 74.7%
- 5
Gemini 2.5 Flash Sep 2025 ↓1 74.4%
- 6
DeepSeek-R1 May 2025 ↑1 71.9%
- 7 gpt-oss-20b ↑2 71.8%
- 8 GPT-5 nano 71.8%
- 9
Gemini 2.5 Flash-Lite Sep 2025 ↑1 71.8%
- 10 GPT-5 ↓4 71.3%
Logical Process
13 models we track have a score here.
- 1
Qwen3-235B-A22B-Thinking Jul 2025 72.8%
- 2
Qwen3-30B-A3B-Thinking Jul 2025 ↑1 69.6%
- 3
Gemini 2.5 Flash Sep 2025 ↑1 68.2%
- 4 GPT-5 mini ↓2 67.4%
- 5
DeepSeek-V3.1 66.7%
- 6
DeepSeek-R1 May 2025 ↑1 66.5%
- 7
Gemini 2.5 Flash-Lite Sep 2025 ↑3 65.8%
- 8 GPT-5 ↓2 65.2%
- 9 gpt-oss-20b 64.2%
- 10 GPT-5 nano ↓2 63.8%
Helpful Outcome
13 models we track have a score here.
- 1 GPT-5 mini ↑1 80.8%
- 2 GPT-5 ↑4 77.8%
- 3 GPT-5 nano ↑5 76.6%
- 4
Qwen3-235B-A22B-Thinking Jul 2025 ↓3 75.4%
- 5
Qwen3-30B-A3B-Thinking Jul 2025 ↓2 74.7%
- 6
Gemini 2.5 Flash Sep 2025 ↓2 74.2%
- 7
DeepSeek-R1 May 2025 73.1%
- 8 gpt-oss-20b ↑1 73.1%
- 9
DeepSeek-V3.1 ↓4 72%
- 10 gpt-oss-120b ↑1 71%
Harmless Outcome
13 models we track have a score here.
- 1 GPT-5 mini ↑1 88.2%
- 2 GPT-5 nano ↑6 87.5%
- 3 GPT-5 ↑3 87.2%
- 4
Qwen3-235B-A22B-Thinking Jul 2025 ↓3 85.5%
- 5 gpt-oss-120b ↑6 85.5%
- 6
Gemini 2.5 Flash Sep 2025 ↓2 85.5%
- 7 gpt-oss-20b ↑2 84.9%
- 8 Claude Sonnet 4 ↑5 84.8%
- 9
Qwen3-30B-A3B-Thinking Jul 2025 ↓6 84.2%
- 10
DeepSeek-V3.1 ↓5 84.2%
More effort adds a median 12 score
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: The MoReBench paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Final response | 77.6 | ||||
| GPT-5 mini | Final response, high reasoning effort | 75.6 | |||
| Final response | 75.4 | ||||
| Final response | 75.0 | ||||
| Final response | 73.9 | ||||
| GPT-5 | Final response, high reasoning effort | 73.2 | |||
| Final response | 72.9 | ||||
| GPT-5 nano | Final response, high reasoning effort | 72.3 | |||
| gpt-oss-20b | Final response | 71.7 | |||
| Final response | 71.7 | ||||
| Thinking trace | 70.3 | ||||
| gpt-oss-120b | Final response | 69.9 | |||
| Thinking trace | 69.8 | ||||
| Claude Opus 4.1 | Final response | 65.5 | |||
| Thinking trace | 64.5 | ||||
| GPT-5 mini | Thinking trace, high reasoning effort | 64.0 | |||
| Claude Sonnet 4 | Final response | 63.4 | |||
| GPT-5 | Thinking trace, high reasoning effort | 61.3 | |||
| GPT-5 nano | Thinking trace, high reasoning effort | 60.9 | |||
| gpt-oss-20b | Thinking trace | 57.7 | |||
| Claude Sonnet 4 | Thinking trace | 57.1 | |||
| Thinking trace | 53.0 | ||||
| Claude Opus 4.1 | Thinking trace | 50.9 | |||
| gpt-oss-120b | Thinking trace | 50.8 | |||
| Thinking trace | 41.4 | ||||
| Thinking trace | 36.6 |
More effort adds a median 12 score
Models the leaderboard ran at two or more effort levels. We don’t have enough cost data here to say what the extra effort costs.
Source: The MoReBench paper
| Model | Effort | Harness | Score | Cost per task | Output tokens |
|---|---|---|---|---|---|
| Final response | 77.6 | ||||
| GPT-5 mini | Final response, high reasoning effort | 75.6 | |||
| Final response | 75.4 | ||||
| Final response | 75.0 | ||||
| Final response | 73.9 | ||||
| GPT-5 | Final response, high reasoning effort | 73.2 | |||
| Final response | 72.9 | ||||
| GPT-5 nano | Final response, high reasoning effort | 72.3 | |||
| gpt-oss-20b | Final response | 71.7 | |||
| Final response | 71.7 | ||||
| Thinking trace | 70.3 | ||||
| gpt-oss-120b | Final response | 69.9 | |||
| Thinking trace | 69.8 | ||||
| Claude Opus 4.1 | Final response | 65.5 | |||
| Thinking trace | 64.5 | ||||
| GPT-5 mini | Thinking trace, high reasoning effort | 64.0 | |||
| Claude Sonnet 4 | Final response | 63.4 | |||
| GPT-5 | Thinking trace, high reasoning effort | 61.3 | |||
| GPT-5 nano | Thinking trace, high reasoning effort | 60.9 | |||
| gpt-oss-20b | Thinking trace | 57.7 | |||
| Claude Sonnet 4 | Thinking trace | 57.1 | |||
| Thinking trace | 53.0 | ||||
| Claude Opus 4.1 | Thinking trace | 50.9 | |||
| gpt-oss-120b | Thinking trace | 50.8 | |||
| Thinking trace | 41.4 | ||||
| Thinking trace | 36.6 |
Number one over time
Lab claims count from the day the card came out, independent runs from the model’s release.
Source: The MoReBench paper
Scores over time
All the scores we have, one dot each, so a model tested at several settings appears more than once.
Source: Official MoReBench leaderboard
No lab model card reports it yet
We read the headline results of 540 launch posts from 57 labs, and none of them list this benchmark.
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%
- 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%
- Covert misbehavior (Thinking Machines) In use Agentic tasks that check whether a model misbehaves and conceals it. Scored by the rate of covert misbehavior.
Inkling 2 Mini 8.30%
- Destructive actions (Thinking Machines) In use Agentic tasks that check whether a model takes destructive actions it was not authorized to take. Scored by the rate of destructive actions.
Inkling 2 Small 4.70%
- Misbehavior (Thinking Machines) In use Agentic tasks that check whether a model fabricates results, ignores instructions or hides what it did when a task is hard. Scored by the rate of misbehavior.
Inkling 2 Mini 9.80%
Data audit
8 of 9 checks pass, 1 warning. Look first at the Newest score check. The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash-Lite. Every benchmark
Catalog record 2 pass
- Pass Task access All tasks are public.
- Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
- Pass Independent runs Scores from the benchmark’s paper (14 entries).
- Pass Models scored 13 models we track, plus 1 other entry on its leaderboard.
- Warning Newest score The newest came 14 months ago, in September 2025, for Gemini 2.5 Flash-Lite.
- Note Frontier models None of the top 10 has a score. The strongest tested is GPT-5 (#71).
- Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
- Pass Importer We read the board on arxiv.org: 26 rows for models we track and 1 other entry at the last read.
- Pass Last read November 20th, 2026.
- Pass Read worked The last read worked.
Consistency 1 pass 1 note
- Note Lab claims No model has both a lab number and an independent run.
- Pass Known issues No published audit found problems.