whatmodel.sh / Koala 13B
Koala 13B
↗ share
+ contribute
Koala 13B is an open-weights language model from UC Berkeley (BAIR), released April 2023. Its published size is 13B parameters. Licence: koala-academic-research-only.
Catalogue identity notes
Comparison-only. Arena entry 'koala-13b' (text). LMArena's leaderboard table (arena-leaderboard space @6f2eeeb) labels 'koala-13b' as 'Koala-13B' (UC Berkeley, Non-commercial), linking https://bair.berkeley.edu/blog/2023/04/03/koala/. Maker repo young-geng/koala @ea9d262: licence tag other. Parameters: nominal size from the model name. Released 2023-04: BAIR blog post dated 2023-04-03 (linked by LMArena's leaderboard table). Weights were published as a diff against LLaMA 13B under LLaMA's model licence (young-geng/koala); no single-command run option.
Model information sources
$ how it compares
The LLMs ranked either side of Koala 13B, out of 828 with a score.
| # | model | score | where |
|---|---|---|---|
| 756 |  GPT-NeoX-20B | 28±1 | local |
| 757 |  LLaMA 13B | 28±1 | comparison only |
| 758 | Koala 13B | 28±1 | comparison only |
| 759 |  StarCoderBase 3B | 28±1 | local |
| 760 |  Granite 3.0 1B-A400M Instruct | 27±1 | local |
$ what it's good at
coding▁▁▃▃▅▇█▇▅▄▃▁ #509 of 524 * · 19% of its score · 1 test
Beats Molmo 7B-D (0924), DeepSeek R1 Distill Qwen 1.5B, Qwen3 0.6B here, though they rank higher overall. Trails Falcon3 1B Instruct, Gemma 2B IT, Granite 3.0 1B-A400M Instruct here, though they rank lower overall.
chat & writing▁▁▃▃▅▄▆█▇▇▇▃ #446 of 463 · 50% of its score · 3 tests
Beats Phi-2 (2.7B), Falcon 40B Instruct here, though they rank higher overall. Trails Gemma 3 270M IT, SmolLM2 360M Instruct, Granite 4.0 350M here, though they rank lower overall.
| test | result | vs the others in its run | counts |
|---|---|---|---|
| Polyrating rating (bias-aware re-fit of Chatbot Arena votes, mid-2024) | 967.0 ✎ | ▂▁▂▃▃▇█▆▇▄▂▂ beats 11/95 | 17% |
| LMArena (text) | 1070.9 ✎ | ▁▁▃▂▃▃▃▆▆▇█▃ beats 12/379 | 17% |
| MT-Bench (GPT-4 single-answer grading, LMSYS runs) | 5.3500 MT-Bench score (1-10) ✎ | ▆▂▂▆▆▄██▂▄▄▆ beats 10/28 | 15% |
other tests3 not in a task · 31% of its score
| test | result | vs the others in its run | counts |
|---|---|---|---|
| BIG-Bench Hard | 34.6% ✎ | ███··█·█·█·█ beats 3/6 | 17% |
| MMLU | 44.6% ✎ | ▄▄·▄···▄▄·██ beats 5/8 | 7.2% |
| EQ-Bench v1 (60 questions, paper run) | 24.9% ✎ | ▂··▄·▂▆▄▂█·▂ beats 1/14 | 6.1% |
$ where to run it
No verified API or download recorded yet · add one
Weights and documentation: huggingface.co/young-geng ↗
$ how the score was worked out · InstructEval (DeCLaRe Lab) RE… 43%, Polyrating: A Cost-Effective… 17%, LMArena leaderboard dataset 17% and 2 more
Each result implies a score on its own, after adjusting for how hard its test is and who ran it. The score is their weighted average, with a small pull toward the middle for thinly tested models. Each test's weight is its counts column above; here it's added up by who ran the tests. How it works
| source | results | share of the score |
|---|---|---|
 InstructEval (DeCLaRe Lab) README results table | 3 | 43% |
 Polyrating: A Cost-Effective and Bias-Aware Rating System for LLM Evaluation (arXiv:2409.00696v3) | 1 | 17% |
 LMArena leaderboard dataset | 1 | 17% |
 LMSYS Chatbot Arena leaderboard Space: MT-Bench column of the published leaderboard table | 1 | 15% |
| EQ-Bench: An Emotional Intelligence Benchmark for Large Language Models — authors' own runs | 1 | 6% |
| prior · pull toward the middle | 1% | |
| weighted average | = 27.7 |
$ improve this page
Use a short GitHub form or edit the data through a pull request. Every number needs a source; here's which ones we can use.
data updated · sources and licences · json