Benchmark Almanac

AGI-Eval Community Base Language Models

Kimi K2 Sep 2025 0.71 leads since September 2025
Not in model cards

Base models answer Chinese and English questions across seven skill areas. The board reports objective accuracy.

The 2026/11 leaderboard reports 25 scored model entries on its primary measure.

Released
February 2025
Kind
Reasoning in many languages
Level
School
Scored
Checked automatically
Answers
Multiple choice
Tasks
Kept private
Measures
Capability
Language
Chinese, English
In and out
Text → Text
Interaction
Single turn
Refresh
One fixed set
Run by
A host
Strongest model tested
DeepSeek-V3.2-Exp #99 on the capability index
Last model added
September 2025
Scores refreshed
November 21st, 2026
25
Entries on its own leaderboard, 23 models we track

Kimi K2 leads with 0.71 among models released by September 2025 Kimi K2 leads with 0.71 among models released by September 2025

Best score per entry from the benchmark's own leaderboard, read on November 21st, 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. 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 Kimi K2 Sep 2025 0.71
  2. 2 DeepSeek-V3.1 0.68
  3. 3 DeepSeek-V3.2-Exp 0.67
  4. 4 DeepSeek-V3 Mar 2025 0.67
  5. 5 ERNIE-4.5-300B-A47B 0.66
  6. 6 Llama4-Maverick-17B-128E-基座 0.65
  7. 7 Qwen2.5-32B 0.63
  8. 8 Qwen2.5-72B 0.63
  9. 9 Qwen3-30B-A3B 0.63
  10. 10 Qwen3-14B 0.61
  11. 11 GLM-4.5 0.61
  12. 12 Qwen3-8B 0.58
  13. 13 Llama 3.1-405B 0.58
  14. 14 Qwen2.5-14B 0.57
  15. 15 Qwen3-4B 0.55
  16. 16 Qwen2.5-7B 0.54
  17. 17 Gemma 3 27B 0.53
  18. 18 Llama 3.1-70B 0.53
  19. 19 Llama4-Scout-17B-16E-基座 0.52
  20. 20 GLM-4-9B 0.51
  21. 21 Gemma 3 12B 0.49
  22. 22 Qwen3-1.7B 0.45
  23. 23 Gemma 3 4B 0.36
  24. 24 Qwen3-0.6B 0.36
  25. 25 Gemma 3 1B 0.19

Source: Official AGI-Eval Community Base Language Models leaderboard

#EntryScoreSettingReleasedRun onSource
1 Kimi K2 Sep 2025 0.71 Sep 2025 Dec 2025 Official leaderboard
2 DeepSeek-V3.1 0.68 Aug 2025 Mar 2026 Official leaderboard
3 DeepSeek-V3.2-Exp 0.67 Sep 2025 Mar 2026 Official leaderboard
4 DeepSeek-V3 Mar 2025 0.67 Mar 2025 Jun 2026 Official leaderboard
5 ERNIE-4.5-300B-A47B 0.66 Jun 2025 Aug 2025 Official leaderboard
6 Llama4-Maverick-17B-128E-基座 0.65 Mar 2026 Official leaderboard
7 Qwen2.5-32B 0.63 Sep 2024 Jun 2026 Official leaderboard
8 Qwen2.5-72B 0.63 Sep 2024 Jun 2026 Official leaderboard
9 Qwen3-30B-A3B 0.63 Apr 2025 Aug 2025 Official leaderboard
10 Qwen3-14B 0.61 Apr 2025 Aug 2025 Official leaderboard
11 GLM-4.5 0.61 Aug 2025 Dec 2025 Official leaderboard
12 Qwen3-8B 0.58 Apr 2025 Mar 2026 Official leaderboard
13 Llama 3.1-405B 0.58 Jul 2024 Jun 2026 Official leaderboard
14 Qwen2.5-14B 0.57 Sep 2024 Jun 2026 Official leaderboard
15 Qwen3-4B 0.55 Apr 2025 Feb 2026 Official leaderboard
16 Qwen2.5-7B 0.54 Sep 2024 Mar 2026 Official leaderboard
17 Gemma 3 27B 0.53 Mar 2025 Dec 2025 Official leaderboard
18 Llama 3.1-70B 0.53 Jul 2024 Dec 2025 Official leaderboard
19 Llama4-Scout-17B-16E-基座 0.52 Feb 2026 Official leaderboard
20 GLM-4-9B 0.51 Jun 2024 Mar 2026 Official leaderboard
21 Gemma 3 12B 0.49 Mar 2025 Jun 2025 Official leaderboard
22 Qwen3-1.7B 0.45 Apr 2025 Mar 2026 Official leaderboard
23 Gemma 3 4B 0.36 Mar 2025 Mar 2026 Official leaderboard
24 Qwen3-0.6B 0.36 Apr 2025 Dec 2025 Official leaderboard
25 Gemma 3 1B 0.19 Mar 2025 Mar 2026 Official leaderboard
#ModelScoreSettingReleasedRun onSource
1 Kimi K2 Sep 2025 0.71 Sep 2025 Dec 2025 Official leaderboard
2 DeepSeek-V3.1 0.68 Aug 2025 Mar 2026 Official leaderboard
3 DeepSeek-V3.2-Exp 0.67 Sep 2025 Mar 2026 Official leaderboard
4 DeepSeek-V3 Mar 2025 0.67 Mar 2025 Jun 2026 Official leaderboard
5 ERNIE-4.5-300B-A47B 0.66 Jun 2025 Aug 2025 Official leaderboard
6 Qwen2.5-32B 0.63 Sep 2024 Jun 2026 Official leaderboard
7 Qwen2.5-72B 0.63 Sep 2024 Jun 2026 Official leaderboard
8 Qwen3-30B-A3B 0.63 Apr 2025 Aug 2025 Official leaderboard
9 Qwen3-14B 0.61 Apr 2025 Aug 2025 Official leaderboard
10 GLM-4.5 0.61 Aug 2025 Dec 2025 Official leaderboard
11 Qwen3-8B 0.58 Apr 2025 Mar 2026 Official leaderboard
12 Llama 3.1-405B 0.58 Jul 2024 Jun 2026 Official leaderboard
13 Qwen2.5-14B 0.57 Sep 2024 Jun 2026 Official leaderboard
14 Qwen3-4B 0.55 Apr 2025 Feb 2026 Official leaderboard
15 Qwen2.5-7B 0.54 Sep 2024 Mar 2026 Official leaderboard
16 Gemma 3 27B 0.53 Mar 2025 Dec 2025 Official leaderboard
17 Llama 3.1-70B 0.53 Jul 2024 Dec 2025 Official leaderboard
18 GLM-4-9B 0.51 Jun 2024 Mar 2026 Official leaderboard
19 Gemma 3 12B 0.49 Mar 2025 Jun 2025 Official leaderboard
20 Qwen3-1.7B 0.45 Apr 2025 Mar 2026 Official leaderboard
21 Gemma 3 4B 0.36 Mar 2025 Mar 2026 Official leaderboard
22 Qwen3-0.6B 0.36 Apr 2025 Dec 2025 Official leaderboard
23 Gemma 3 1B 0.19 Mar 2025 Mar 2026 Official leaderboard

Rankings by part

We read 9 parts from the benchmark’s own leaderboard in 2 groups: skill and language. Each ranks models on one part of the test. Kimi K2 leads 5 of them, and 3 other models lead the rest. Arrows show how far a model moved from its overall rank among the models we track.

Code reasoning

23 models we track have a score here.

  1. 1 Kimi K2 Sep 2025 0.52
  2. 2 DeepSeek-V3.1 0.51
  3. 3 Qwen3-30B-A3B ↑5 0.50
  4. 4 DeepSeek-V3.2-Exp ↓1 0.50
  5. 5 DeepSeek-V3 Mar 2025 ↓1 0.50
  6. 6 Qwen3-14B ↑3 0.48
  7. 7 Qwen2.5-72B 0.46
  8. 8 Qwen2.5-32B ↓2 0.45
  9. 9 Qwen3-8B ↑2 0.45
  10. 10 Llama 3.1-405B ↑2 0.42

4 models have held number one on the official leaderboard since its release in February 2025

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

Qwen DeepSeek Moonshot AI
Own leaderboard Not tested since Sep 2025
Kimi K2 0.71

Source: Official AGI-Eval Community Base Language Models leaderboard

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
On the benchmark’s own leaderboard
0.00 0.20 0.40 0.60 0.80 2024 2025 0.00 0.20 0.40 0.60 0.80 2024 2025

Source: Official AGI-Eval Community Base Language Models 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 Multilingual

All 545 in Multilingual
  • MMMLU In use MMLU professionally translated into 14 languages. Multiple-choice accuracy. Claude Mythos Preview 92.7%
  • WMT24++ In use English-to-55-language translation of WMT24 test data. Scored with automatic metrics such as MetricX or xCOMET. K-EXAONE 90.5%
  • MMLU-ProX In use MMLU-Pro translated into up to 29 languages with expert review. Multiple-choice accuracy per language. Qwen3.7-Max 87%
  • CLIcK In use Models answer 1,995 multiple-choice questions on Korean language and culture. Questions come from official Korean exams and textbooks in eleven categories. Solar Open 2 90.7%
  • Global MMLU Lite 3.0 In use Models answer 400 human-translated MMLU questions per language in 23 languages. Version 3.0 adds Czech, Hungarian, Oriya, Slovak and Tajik and updates Italian. Accuracy. Claude Opus 5 95.7%
  • MMMB In use Models answer multiple-choice questions about images in six languages across 15 categories. The set has about 12,000 questions. LFM2.5-VL-3B 83%

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 DeepSeek-V3.2-Exp. Every benchmark

Catalog record 2 pass
  • Pass Task access The tasks are private.
  • Pass Confidence Checked against a primary source.
Scores 1 warning 2 pass 2 notes
  • Pass Independent runs Scores from the official leaderboard (25 entries).
  • Pass Models scored 23 models we track, plus 2 other entries on its leaderboard.
  • Warning Newest score The newest came 14 months ago, in September 2025, for DeepSeek-V3.2-Exp.
  • Note Frontier models None of the top 10 has a score. The strongest tested is DeepSeek-V3.2-Exp (#99).
  • Note Headroom Scored as a plain number, with no fixed ceiling.
Leaderboard import 3 pass
  • Pass Importer We read the board on agi-eval.org: 23 rows for models we track and 2 other entries at the last read.
  • Pass Last read November 21st, 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