Benchmark Almanac

CaBBQ

Qwen2.5-7B 81% leads since July 2025
Not in model cards

Models answer Catalan questions about people in ambiguous and unambiguous situations. The test measures whether answers follow social stereotypes when evidence is insufficient.

Released
July 2025
Kind
Bias, fairness and politics
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Propensity
Risk
Hate
Behavior
Bias
Language
Catalan
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
In views
Content safety
Strongest model tested
Llama 3.1-8B #203 on the capability index
Last model added
March 2025
Scores refreshed
November 20th, 2026
34
Entries in its paper, 11 models we track

Qwen2.5-7B leads with 81% among models released by March 2025 Qwen2.5-7B leads with 81% among models released by March 2025

Best score per entry from the benchmark's paper, read on November 20th, 2026. A paper’s table doesn’t add newer models. The paper reports no overall accuracy, so this is the plain mean of ambiguous and disambiguated context accuracy. 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. This view shows only models we track, each at its best.

The newest model with a score here came out in March 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 Qwen2.5-7B Zero-shot 81%
  2. 2 Gemma 3 12B Zero-shot 79.5%
  3. 3 Mistral-v0.3 7B instruct Zero-shot 75.5%
  4. 4 Qwen2.5-3B Zero-shot 75.5%
  5. 5 Llama 3.1-8B Zero-shot 65%
  6. 6 Gemma 3 4B Zero-shot 63.5%
  7. 7 Occiglot-EU5 7B instruct Zero-shot 62%
  8. 8 EuroLLM 9B instruct Zero-shot 61.5%
  9. 9 Salamandra 7B instruct Zero-shot 51%
  10. 10 Gemma 3 12B PT Zero-shot 50.5%
  11. 11 Gemma 3 1B Zero-shot 50%
  12. 12 Llama-3.1 8B base Zero-shot 48%
  13. 13 EuroLLM 9B base Zero-shot 47.5%
  14. 14 Qwen2.5 7B base Zero-shot 46.5%
  15. 15 Mistral 7B v0.3 Zero-shot 46%
  16. 16 Salamandra 7B base Zero-shot 45.5%
  17. 17 Qwen2.5-1.5B Zero-shot 45%
  18. 18 Gemma 3 4B PT Zero-shot 45%
  19. 19 Occiglot-EU5 7B base Zero-shot 44.5%
  20. 20 FLOR 1.3B base Zero-shot 44%
  21. 21 Qwen2.5 1.5B base Zero-shot 44%
  22. 22 Qwen2.5 3B base Zero-shot 44%
  23. 23 Tower-v0.1 7B instruct Zero-shot 43%
  24. 24 FLOR 1.3B instruct Zero-shot 42%
  25. 25 EuroLLM 1.7B base Zero-shot 41.5%
  26. 26 Salamandra 2B base Zero-shot 41.5%
  27. 27 Tower-v0.1 13B instruct Zero-shot 41.5%
  28. 28 Gemma 3 1B PT Zero-shot 41%
  29. 29 Tower-v0.1 13B base Zero-shot 40.5%
  30. 30 FLOR 6.3B base Zero-shot 39.5%
  31. 31 Tower-v0.1 7B base Zero-shot 38.5%
  32. 32 FLOR 6.3B instruct Zero-shot 38%
  33. 33 Salamandra 2B instruct Zero-shot 36%
  34. 34 EuroLLM 1.7B instruct Zero-shot 35.5%

Source: The CaBBQ paper

#EntryScoreSettingDateSource
1 Qwen2.5-7B 81% Zero-shot Sep 2024 Official leaderboard
2 Gemma 3 12B 79.5% Zero-shot Mar 2025 Official leaderboard
3 Mistral-v0.3 7B instruct 7B 75.5% Zero-shot Official leaderboard
4 Qwen2.5-3B 75.5% Zero-shot Sep 2024 Official leaderboard
5 Llama 3.1-8B 65% Zero-shot Jul 2024 Official leaderboard
6 Gemma 3 4B 63.5% Zero-shot Mar 2025 Official leaderboard
7 Occiglot-EU5 7B instruct 7B 62% Zero-shot Official leaderboard
8 EuroLLM 9B instruct 9B 61.5% Zero-shot Official leaderboard
9 Salamandra 7B instruct 7B 51% Zero-shot Official leaderboard
10 Gemma 3 12B PT 50.5% Zero-shot Mar 2025 Official leaderboard
11 Gemma 3 1B 50% Zero-shot Mar 2025 Official leaderboard
12 Llama-3.1 8B base 8B 48% Zero-shot Official leaderboard
13 EuroLLM 9B base 9B 47.5% Zero-shot Official leaderboard
14 Qwen2.5 7B base 7B 46.5% Zero-shot Official leaderboard
15 Mistral 7B v0.3 46% Zero-shot Sep 2023 Official leaderboard
16 Salamandra 7B base 7B 45.5% Zero-shot Official leaderboard
17 Qwen2.5-1.5B 45% Zero-shot Sep 2024 Official leaderboard
18 Gemma 3 4B PT 45% Zero-shot Mar 2025 Official leaderboard
19 Occiglot-EU5 7B base 7B 44.5% Zero-shot Official leaderboard
20 FLOR 1.3B base 1.3B 44% Zero-shot Official leaderboard
21 Qwen2.5 1.5B base 1.5B 44% Zero-shot Official leaderboard
22 Qwen2.5 3B base 3B 44% Zero-shot Official leaderboard
23 Tower-v0.1 7B instruct 7B 43% Zero-shot Official leaderboard
24 FLOR 1.3B instruct 1.3B 42% Zero-shot Official leaderboard
25 EuroLLM 1.7B base 1.7B 41.5% Zero-shot Official leaderboard
26 Salamandra 2B base 2B 41.5% Zero-shot Official leaderboard
27 Tower-v0.1 13B instruct 13B 41.5% Zero-shot Official leaderboard
28 Gemma 3 1B PT 41% Zero-shot Mar 2025 Official leaderboard
29 Tower-v0.1 13B base 13B 40.5% Zero-shot Official leaderboard
30 FLOR 6.3B base 6.3B 39.5% Zero-shot Official leaderboard
31 Tower-v0.1 7B base 7B 38.5% Zero-shot Official leaderboard
32 FLOR 6.3B instruct 6.3B 38% Zero-shot Official leaderboard
33 Salamandra 2B instruct 2B 36% Zero-shot Official leaderboard
34 EuroLLM 1.7B instruct 1.7B 35.5% Zero-shot Official leaderboard
#ModelScoreSettingDateSource
1 Qwen2.5-7B 81% Zero-shot Sep 2024 Official leaderboard
2 Gemma 3 12B 79.5% Zero-shot Mar 2025 Official leaderboard
3 Qwen2.5-3B 75.5% Zero-shot Sep 2024 Official leaderboard
4 Llama 3.1-8B 65% Zero-shot Jul 2024 Official leaderboard
5 Gemma 3 4B 63.5% Zero-shot Mar 2025 Official leaderboard
6 Gemma 3 12B PT 50.5% Zero-shot Mar 2025 Official leaderboard
7 Gemma 3 1B 50% Zero-shot Mar 2025 Official leaderboard
8 Mistral 7B v0.3 46% Zero-shot Sep 2023 Official leaderboard
9 Qwen2.5-1.5B 45% Zero-shot Sep 2024 Official leaderboard
10 Gemma 3 4B PT 45% Zero-shot Mar 2025 Official leaderboard
11 Gemma 3 1B PT 41% Zero-shot Mar 2025 Official leaderboard

Rankings by part

We read 2 parts from the benchmark’s paper: Ambiguous context accuracy and Disambiguated context accuracy. Each ranks models on one part of the test. Qwen2.5-7B leads Ambiguous context accuracy and Llama 3.1-8B leads Disambiguated context accuracy. Arrows show how far a model moved from its overall rank among the models we track.

Ambiguous context accuracy

11 models we track have a score here.

  1. 1 Qwen2.5-7B 91%
  2. 2 Qwen2.5-3B ↑1 88%
  3. 3 Gemma 3 12B ↓1 74%
  4. 4 Gemma 3 1B ↑3 51%
  5. 5 Gemma 3 4B 50%
  6. 6 Llama 3.1-8B ↓2 44%
  7. 7 Gemma 3 1B PT ↑4 32%
  8. 8 Qwen2.5-1.5B ↑1 25%
  9. 9 Gemma 3 4B PT ↑1 22%
  10. 10 Gemma 3 12B PT ↓4 17%

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 Jul 2025

Source: The CaBBQ 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 Best so far
0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Qwen2.5-7B Llama 3.1-8B Mistral 7B v0.3 0% 20% 40% 60% 80% 100% 2024 2025 Near the ceiling (85%) Qwen2.5-7B Llama 3.1-8B

Source: Official CaBBQ 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

9 of 10 checks pass, 1 warning. Look first at the Newest score check. The newest came 20 months ago, in March 2025, for Gemma 3 12B. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 3 pass 1 note
  • Pass Independent runs Scores from the benchmark’s paper (34 entries).
  • Pass Models scored 11 models we track, plus 23 other entries on its leaderboard.
  • Warning Newest score The newest came 20 months ago, in March 2025, for Gemma 3 12B.
  • Note Frontier models None of the top 10 has a score. The strongest tested is Llama 3.1-8B (#203).
  • Pass Headroom The best is 81%, under 85%, so it still separates models.
Leaderboard import 3 pass
  • Pass Importer We read the board on arxiv.org: 11 rows for models we track and 23 other entries 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