<picture> <source media="(prefers-color-scheme: dark)" srcset="assets/logo/kanana-logo-light.png"> <source media="(prefers-color-scheme: light)" srcset="assets/logo/kanana-logo-dark.png"> <img alt="Kanana Logo" src="assets/logo/kanana-logo-light.png" width="400" style="margin: 40px auto;"> </picture> 🤗 <a href="https://kko.kakao.com/kananallm">Kanana-1.5 HF Models</a> &nbsp | &nbsp 📕 <a href="https://tech.kakao.com/search?q=kanana">Blogs</a> &nbsp

News 🔥

✨2025/07/24: Published a blog post about Kanana-1.5-15.7B-A3B models and released 🤗HF model weights.
📕2025/05/23: Published a blog post about Kanana 1.5 models and released 🤗HF model weights.
📜2025/02/27: Released Technical Report and 🤗HF model weights.
📕2025/01/10: Published a blog post about the development of Kanana Nano model.
📕2024/11/14: Published blog posts (pre-training, post-training) about the development of Kanana models.
▶️2024/11/06: Published a presentation video about the development of the Kanana models.

Kanana 1.5
Kanana 1.0
Citation
Contact

Kanana 1.5

Kanana 1.5, a newly introduced version of the Kanana model family, presents substantial enhancements in coding, mathematics, and function calling capabilities over the previous version, enabling broader application to more complex real-world problems. This new version now can handle up to 32K tokens length natively and up to 128K tokens using YaRN, allowing the model to maintain coherence when handling extensive documents or engaging in extended conversations. Furthermore, Kanana 1.5 delivers more natural and accurate conversations through a refined post-training process.

✨ Updated Kanana-1.5-15.7B-A3B

Introducing Kanana-1.5-15.7B-A3B, the first Mixture-of-Experts (MoE) model in our Kanana family, engineered for exceptional efficiency and powerful performance. Kanana-1.5-15.7B-A3B, which has sparse architecture, delivers capabilities comparable to the Kanana-1.5-8B dense model while utilizing only 37% of the FLOPS per token, making it a highly inference-efficient and cost-effective solution for real-world applications. Furthermore, Kanana-1.5-15.7B-A3B is powered by our newly enhanced post-training strategy, which includes on-policy distillation followed by reinforcement learning.

[!Note] Neither the pre-training nor the post-training data includes Kakao user data.

Performance

Base Model Evaluation

<table> <tr> <th>Models</th> <th>MMLU</th> <th>KMMLU</th> <th>HAE-RAE</th> <th>HumanEval</th> <th>MBPP</th> <th>GSM8K</th> </tr> <tr> <td>Kanana-Flag-1.5-32.5B</td> <td align="center">76.76</td> <td align="center">61.90</td> <td align="center">89.18</td> <td align="center">73.17</td> <td align="center">65.60</td> <td align="center">81.50</td> </tr> <tr> <td>Kanana-Flag-32.5B</td> <td align="center">77.68</td> <td align="center">62.10</td> <td align="center">90.47</td> <td align="center">51.22</td> <td align="center">63.40</td> <td align="center">70.05</td> </tr> <tr> <td>Kanana-Essence-1.5-9.8B</td> <td align="center">68.27</td> <td align="center">52.78</td> <td align="center">86.34</td> <td align="center">64.63</td> <td align="center">61.60</td> <td align="center">71.57</td> </tr> <tr> <td>Kanana-Essence-9.8B</td> <td align="center">67.61</td> <td align="center">50.57</td> <td align="center">84.97</td> <td align="center">40.24</td> <td align="center">53.60</td> <td align="center">63.61</td> </tr>  <tr> <td>Kanana-Nano-1.5-3B</td> <td align="center">59.23</td> <td align="center">47.30</td> <td align="center">78.00</td> <td align="center">46.34</td> <td align="center">46.80</td> <td align="center">61.79</td> </tr>  <tr> <td>Kanana-Nano-2.1B</td> <td align="center">54.83</td> <td align="center">44.80</td> <td align="center">77.09</td> <td align="center">31.10</td> <td align="center">46.20</td> <td align="center">46.32</td> </tr> </table>

Open-source Base Model Evaluation

<table> <tr> <th>Models</th> <th>MMLU</th> <th>KMMLU</th> <th>HAE-RAE</th> <th>HumanEval</th> <th>MBPP</th> <th>GSM8K</th> </tr> <tr> <td>Kanana-1.5-15.7B-A3B (New!)</td> <td align="center">64.79</td> <td align="center">51.77</td> <td align="center">83.23</td> <td align="center">59.76</td> <td align="center">60.10</td> <td align="center">61.18</td> </tr> <tr> <td>Kanana-1.5-8B</td> <td align="center">64.24</td> <td align="center">48.94</td> <td align="center">82.77</td> <td align="center">61.59</td> <td align="center">57.80</td> <td align="center">63.53</td> </tr> <tr> <td>Kanana-8B*</td> <td align="center">64.22</td> <td align="center">48.30</td> <td align="center">83.41</td> <td align="center">40.24</td> <td align="center">51.40</td> <td align="center">57.09</td> </tr> <tr> <td>Kanana-1.5-2.1B</td> <td align="center">56.30</td> <td align="center">45.10</td> <td align="center">77.46</td> <td align="center">52.44</td> <td align="center">47.00</td> <td align="center">55.95</td> </tr> <tr> <td>Kanana-Nano-2.1B</td> <td align="center">54.83</td> <td align="center">44.80</td> <td align="center">77.09</td> <td align="center">31.10</td> <td align="center">46.20</td> <td align="center">46.32</td> </tr> </table>

* This model is not an open-sourced, just for comparison with Kanana-1.5-8B

Instruct Model Evaluation

<table> <tr> <th>Models</th> <th>MT-Bench</th> <th>KoMT-Bench</th> <th>IFEval</th> <th>HumanEval+</th> <th>MBPP+</th> <th>GSM8K (0-shot)</th> <th>MATH</th> <th>MMLU (0-shot, CoT)</th> <th>KMMLU (0-shot, CoT)</th> <th>FunctionChatBench</th> </tr> <tr> <td>Kanana-Flag-1.5-32.5B</td> <td align="center">8.13</td> <td align="center">8.12</td> <td align="center">82.70</td> <td align="center">79.88</td> <td align="center">71.96</td> <td align="center">93.03</td> <td align="center">75.96</td> <td align="center">82.76</td> <td align="center">64.10</td> <td align="center">67.17</td> </tr> <tr> <td>Kanana-Flag-32.5B</td> <td align="center">8.33</td> <td align="center">8.03</td> <td align="center">84.59</td> <td align="center">78.66</td> <td align="center">69.84</td> <td align="center">91.66</td> <td align="center">58.08</td> <td align="center">81.08</td> <td align="center">64.19</td> <td align="center">65.67</td> </tr> <tr> <td>Kanana-Essence-1.5-9.8B</td> <td align="center">7.88</td> <td align="center">7.35</td> <td align="center">76.34</td> <td align="center">72.56</td> <td align="center">66.93</td> <td align="center">90.07</td> <td align="center">62.02</td> <td align="center">72.85</td> <td align="center">52.00</td> <td align="center">51.43</td> </tr> <tr> <td>Kanana-E

Kanana

Install / Use

README