Benchmark Almanac

MoReBench

Qwen3-235B-A22B-Thinking Jul 2025 77.6 leads since October 2025
Not in model cards

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.

Released
October 2025
Kind
Deception and alignment
Level
Everyday
Scored
Graded by a model
Answers
Open-ended
Tasks
All public
Measures
Capability
Language
English
In and out
Text → Text
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
14
Entries in its paper, 13 models we track

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:

  1. 1 Qwen3-235B-A22B-Thinking Jul 2025, Final response 77.6
  2. 2 GPT-5 mini Final response, high reasoning effort 75.6
  3. 3 Qwen3-30B-A3B-Thinking Jul 2025, Final response 75.4
  4. 4 Gemini 2.5 Flash Sep 2025, Final response 75.0
  5. 5 Gemini-2.5-Pro (snapshot unspecified) Final response 74.3
  6. 6 DeepSeek-V3.1 Final response 73.9
  7. 7 GPT-5 Final response, high reasoning effort 73.2
  8. 8 DeepSeek-R1 May 2025, Final response 72.9
  9. 9 GPT-5 nano Final response, high reasoning effort 72.3
  10. 10 gpt-oss-20b Final response 71.7
  11. 11 Gemini 2.5 Flash-Lite Sep 2025, Final response 71.7
  12. 12 gpt-oss-120b Final response 69.9
  13. 13 Claude Opus 4.1 Final response 65.5
  14. 14 Claude Sonnet 4 Final response 63.4

Source: The MoReBench paper

#EntryScoreSettingDateSource
1 Qwen3-235B-A22B-Thinking Jul 2025 77.6 Final response Jul 2025 Official leaderboard
2 GPT-5 mini 75.6 Final response, high reasoning effort Aug 2025 Official leaderboard
3 Qwen3-30B-A3B-Thinking Jul 2025 75.4 Final response Jul 2025 Official leaderboard
4 Gemini 2.5 Flash Sep 2025 75.0 Final response Sep 2025 Official leaderboard
5 Gemini-2.5-Pro (snapshot unspecified) 74.3 Final response Official leaderboard
6 DeepSeek-V3.1 73.9 Final response Aug 2025 Official leaderboard
7 GPT-5 73.2 Final response, high reasoning effort Aug 2025 Official leaderboard
8 DeepSeek-R1 May 2025 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 Gemini 2.5 Flash-Lite Sep 2025 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
#ModelScoreSettingDateSource
1 Qwen3-235B-A22B-Thinking Jul 2025 77.6 Final response Jul 2025 Official leaderboard
2 GPT-5 mini 75.6 Final response, high reasoning effort Aug 2025 Official leaderboard
3 Qwen3-30B-A3B-Thinking Jul 2025 75.4 Final response Jul 2025 Official leaderboard
4 Gemini 2.5 Flash Sep 2025 75.0 Final response Sep 2025 Official leaderboard
5 DeepSeek-V3.1 73.9 Final response Aug 2025 Official leaderboard
6 GPT-5 73.2 Final response, high reasoning effort Aug 2025 Official leaderboard
7 DeepSeek-R1 May 2025 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 Gemini 2.5 Flash-Lite Sep 2025 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. 1 Qwen3-235B-A22B-Thinking Jul 2025 76.7
  2. 2 Qwen3-30B-A3B-Thinking Jul 2025 ↑1 74.2
  3. 3 GPT-5 mini ↓1 74.2
  4. 4 Gemini 2.5 Flash Sep 2025 74.2
  5. 5 DeepSeek-V3.1 73.5
  6. 6 DeepSeek-R1 May 2025 ↑1 72.9
  7. 7 gpt-oss-120b ↑4 72.4
  8. 8 GPT-5 ↓2 72.1
  9. 9 Gemini 2.5 Flash-Lite Sep 2025 ↑1 72.0
  10. 10 GPT-5 nano ↓2 71.5

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.

Lowest effort to highest, light to dark
Qwen3-235B-A22B-Thinking
Thinking tr… Final respo…
77.6
GPT-5 mini
Thinking tr… Final respo…
75.6
Qwen3-30B-A3B-Thinking
Thinking tr… Final respo…
75.4
Gemini 2.5 Flash
Thinking tr… Final respo…
75.0
DeepSeek-V3.1
Thinking tr… Final respo…
73.9
GPT-5
Thinking tr… Final respo…
73.2
DeepSeek-R1
Thinking tr… Final respo…
72.9
GPT-5 nano
Thinking tr… Final respo…
72.3
gpt-oss-20b
Thinking tr… Final respo…
71.7
Gemini 2.5 Flash-Lite
Thinking tr… Final respo…
71.7
gpt-oss-120b
Thinking tr… Final respo…
69.9
Claude Opus 4.1
Thinking tr… Final respo…
65.5

Source: The MoReBench paper

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen3-235B-A22B-Thinking Jul 2025 Final response 77.6
GPT-5 mini Final response, high reasoning effort 75.6
Qwen3-30B-A3B-Thinking Jul 2025 Final response 75.4
Gemini 2.5 Flash Sep 2025 Final response 75.0
DeepSeek-V3.1 Final response 73.9
GPT-5 Final response, high reasoning effort 73.2
DeepSeek-R1 May 2025 Final response 72.9
GPT-5 nano Final response, high reasoning effort 72.3
gpt-oss-20b Final response 71.7
Gemini 2.5 Flash-Lite Sep 2025 Final response 71.7
Qwen3-30B-A3B-Thinking Jul 2025 Thinking trace 70.3
gpt-oss-120b Final response 69.9
Qwen3-235B-A22B-Thinking Jul 2025 Thinking trace 69.8
Claude Opus 4.1 Final response 65.5
DeepSeek-R1 May 2025 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
DeepSeek-V3.1 Thinking trace 53.0
Claude Opus 4.1 Thinking trace 50.9
gpt-oss-120b Thinking trace 50.8
Gemini 2.5 Flash Sep 2025 Thinking trace 41.4
Gemini 2.5 Flash-Lite Sep 2025 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.

Lowest effort to highest, light to dark
Qwen3-235B-A22B-Thinking
Thinking tr… Final respo…
77.6
GPT-5 mini
Thinking tr… Final respo…
75.6
Qwen3-30B-A3B-Thinking
Thinking tr… Final respo…
75.4
Gemini 2.5 Flash
Thinking tr… Final respo…
75.0
DeepSeek-V3.1
Thinking tr… Final respo…
73.9
GPT-5
Thinking tr… Final respo…
73.2
DeepSeek-R1
Thinking tr… Final respo…
72.9
GPT-5 nano
Thinking tr… Final respo…
72.3
gpt-oss-20b
Thinking tr… Final respo…
71.7
Gemini 2.5 Flash-Lite
Thinking tr… Final respo…
71.7
gpt-oss-120b
Thinking tr… Final respo…
69.9
Claude Opus 4.1
Thinking tr… Final respo…
65.5

Source: The MoReBench paper

ModelEffortHarnessScoreCost per taskOutput tokens
Qwen3-235B-A22B-Thinking Jul 2025 Final response 77.6
GPT-5 mini Final response, high reasoning effort 75.6
Qwen3-30B-A3B-Thinking Jul 2025 Final response 75.4
Gemini 2.5 Flash Sep 2025 Final response 75.0
DeepSeek-V3.1 Final response 73.9
GPT-5 Final response, high reasoning effort 73.2
DeepSeek-R1 May 2025 Final response 72.9
GPT-5 nano Final response, high reasoning effort 72.3
gpt-oss-20b Final response 71.7
Gemini 2.5 Flash-Lite Sep 2025 Final response 71.7
Qwen3-30B-A3B-Thinking Jul 2025 Thinking trace 70.3
gpt-oss-120b Final response 69.9
Qwen3-235B-A22B-Thinking Jul 2025 Thinking trace 69.8
Claude Opus 4.1 Final response 65.5
DeepSeek-R1 May 2025 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
DeepSeek-V3.1 Thinking trace 53.0
Claude Opus 4.1 Thinking trace 50.9
gpt-oss-120b Thinking trace 50.8
Gemini 2.5 Flash Sep 2025 Thinking trace 41.4
Gemini 2.5 Flash-Lite Sep 2025 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.

Qwen
Benchmark paper Not tested since Oct 2025

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.

Plot with other benchmarks
In the benchmark’s paper
30.0 40.0 50.0 60.0 70.0 80.0 2025 30.0 40.0 50.0 60.0 70.0 80.0 2025

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

All 636 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.

How we check