Benchmark Almanac

EsBBQ

Qwen2.5-7B 85% leads since July 2025
Not in model cards

Models answer Spanish questions about people in ambiguous and unambiguous situations. The evaluation measures how social stereotypes affect their choices.

Released
July 2025
Kind
Bias, fairness and politics
Level
Everyday
Scored
Checked automatically
Answers
Multiple choice
Tasks
All public
Measures
Propensity
Behavior
Bias
Language
Spanish
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
Anyone
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 85% among models released by March 2025 Qwen2.5-7B leads with 85% 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 85%
  2. 2 Gemma 3 12B Zero-shot 81%
  3. 3 Qwen2.5-3B Zero-shot 77%
  4. 4 Mistral-v0.3 7B instruct Zero-shot 76.5%
  5. 5 Gemma 3 4B Zero-shot 69%
  6. 6 Llama 3.1-8B Zero-shot 67.5%
  7. 7 EuroLLM 9B instruct Zero-shot 64.5%
  8. 8 Occiglot-EU5 7B instruct Zero-shot 57%
  9. 9 Qwen2.5-1.5B Zero-shot 55.5%
  10. 10 Gemma 3 1B Zero-shot 53%
  11. 11 Salamandra 7B instruct Zero-shot 50%
  12. 12 Gemma 3 12B PT Zero-shot 47.5%
  13. 13 EuroLLM 9B base Zero-shot 46.5%
  14. 14 Llama-3.1 8B base Zero-shot 46.5%
  15. 15 Qwen2.5 7B base Zero-shot 46%
  16. 16 Occiglot-EU5 7B base Zero-shot 45.5%
  17. 17 Tower-v0.1 13B instruct Zero-shot 45.5%
  18. 18 Qwen2.5 3B base Zero-shot 45%
  19. 19 Mistral 7B v0.3 Zero-shot 45%
  20. 20 EuroLLM 1.7B base Zero-shot 44.5%
  21. 21 Salamandra 7B base Zero-shot 43.5%
  22. 22 Tower-v0.1 13B base Zero-shot 43.5%
  23. 23 Qwen2.5 1.5B base Zero-shot 42.5%
  24. 24 Tower-v0.1 7B instruct Zero-shot 42%
  25. 25 Gemma 3 4B PT Zero-shot 41.5%
  26. 26 FLOR 1.3B base Zero-shot 40%
  27. 27 FLOR 6.3B base Zero-shot 39%
  28. 28 Tower-v0.1 7B base Zero-shot 39%
  29. 29 FLOR 1.3B instruct Zero-shot 38%
  30. 30 Salamandra 2B base Zero-shot 37%
  31. 31 Gemma 3 1B PT Zero-shot 35.5%
  32. 32 Salamandra 2B instruct Zero-shot 34.5%
  33. 33 EuroLLM 1.7B instruct Zero-shot 33.5%
  34. 34 FLOR 6.3B instruct Zero-shot 33.5%

Source: The EsBBQ paper

#EntryScoreSettingDateSource
1 Qwen2.5-7B 85% Zero-shot Sep 2024 Official leaderboard
2 Gemma 3 12B 81% Zero-shot Mar 2025 Official leaderboard
3 Qwen2.5-3B 77% Zero-shot Sep 2024 Official leaderboard
4 Mistral-v0.3 7B instruct 7B 76.5% Zero-shot Official leaderboard
5 Gemma 3 4B 69% Zero-shot Mar 2025 Official leaderboard
6 Llama 3.1-8B 67.5% Zero-shot Jul 2024 Official leaderboard
7 EuroLLM 9B instruct 9B 64.5% Zero-shot Official leaderboard
8 Occiglot-EU5 7B instruct 7B 57% Zero-shot Official leaderboard
9 Qwen2.5-1.5B 55.5% Zero-shot Sep 2024 Official leaderboard
10 Gemma 3 1B 53% Zero-shot Mar 2025 Official leaderboard
11 Salamandra 7B instruct 7B 50% Zero-shot Official leaderboard
12 Gemma 3 12B PT 47.5% Zero-shot Mar 2025 Official leaderboard
13 EuroLLM 9B base 9B 46.5% Zero-shot Official leaderboard
14 Llama-3.1 8B base 8B 46.5% Zero-shot Official leaderboard
15 Qwen2.5 7B base 7B 46% Zero-shot Official leaderboard
16 Occiglot-EU5 7B base 7B 45.5% Zero-shot Official leaderboard
17 Tower-v0.1 13B instruct 13B 45.5% Zero-shot Official leaderboard
18 Qwen2.5 3B base 3B 45% Zero-shot Official leaderboard
19 Mistral 7B v0.3 45% Zero-shot Sep 2023 Official leaderboard
20 EuroLLM 1.7B base 1.7B 44.5% Zero-shot Official leaderboard
21 Salamandra 7B base 7B 43.5% Zero-shot Official leaderboard
22 Tower-v0.1 13B base 13B 43.5% Zero-shot Official leaderboard
23 Qwen2.5 1.5B base 1.5B 42.5% Zero-shot Official leaderboard
24 Tower-v0.1 7B instruct 7B 42% Zero-shot Official leaderboard
25 Gemma 3 4B PT 41.5% Zero-shot Mar 2025 Official leaderboard
26 FLOR 1.3B base 1.3B 40% Zero-shot Official leaderboard
27 FLOR 6.3B base 6.3B 39% Zero-shot Official leaderboard
28 Tower-v0.1 7B base 7B 39% Zero-shot Official leaderboard
29 FLOR 1.3B instruct 1.3B 38% Zero-shot Official leaderboard
30 Salamandra 2B base 2B 37% Zero-shot Official leaderboard
31 Gemma 3 1B PT 35.5% Zero-shot Mar 2025 Official leaderboard
32 Salamandra 2B instruct 2B 34.5% Zero-shot Official leaderboard
33 EuroLLM 1.7B instruct 1.7B 33.5% Zero-shot Official leaderboard
34 FLOR 6.3B instruct 6.3B 33.5% Zero-shot Official leaderboard
#ModelScoreSettingDateSource
1 Qwen2.5-7B 85% Zero-shot Sep 2024 Official leaderboard
2 Gemma 3 12B 81% Zero-shot Mar 2025 Official leaderboard
3 Qwen2.5-3B 77% Zero-shot Sep 2024 Official leaderboard
4 Gemma 3 4B 69% Zero-shot Mar 2025 Official leaderboard
5 Llama 3.1-8B 67.5% Zero-shot Jul 2024 Official leaderboard
6 Qwen2.5-1.5B 55.5% Zero-shot Sep 2024 Official leaderboard
7 Gemma 3 1B 53% Zero-shot Mar 2025 Official leaderboard
8 Gemma 3 12B PT 47.5% Zero-shot Mar 2025 Official leaderboard
9 Mistral 7B v0.3 45% Zero-shot Sep 2023 Official leaderboard
10 Gemma 3 4B PT 41.5% Zero-shot Mar 2025 Official leaderboard
11 Gemma 3 1B PT 35.5% 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 93%
  2. 2 Qwen2.5-3B ↑1 86%
  3. 3 Gemma 3 12B ↓1 79%
  4. 4 Gemma 3 4B 59%
  5. 5 Gemma 3 1B ↑2 54%
  6. 6 Llama 3.1-8B ↓1 52%
  7. 7 Qwen2.5-1.5B ↓1 42%
  8. 8 Gemma 3 1B PT ↑3 20%
  9. 9 Mistral 7B v0.3 14%
  10. 10 Gemma 3 4B PT 14%

Number one over time

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

Qwen
Benchmark paper Passed 85% in Jul 2025, by Qwen2.5-7B

Source: The EsBBQ paper

Qwen2.5-7B on the official leaderboard passed 85% in September 2024

Models released before it came out already scored that high. 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 EsBBQ 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

All 8 checks pass. Every benchmark

Catalog record 2 pass
  • Pass Task access All tasks are public.
  • Pass Confidence Checked against a primary source.
Scores 2 pass 3 notes
  • 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.
  • Note Newest score The newest came 20 months ago, in March 2025, for Gemma 3 12B. That’s expected for a saturated test.
  • Note Frontier models None of the top 10 has a score, which is fine for a retired test. The strongest tested is Llama 3.1-8B (#203).
  • Note Headroom Saturated. The best is 85%, past 85% since September 2024.
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