Japan: AI model benchmarks
AI models developed in Japan. Rank among countries: #8. Strongest model: llm-jp-4.1-33B (Rating 1680, #247 overall). Latest release: llm-jp-4.1-33B (Sep 28, 2026).
Current models, released in the last 18 months
| Rank | Model | Provider | Rating | Released | Access |
|---|---|---|---|---|---|
| 247 | llm-jp-4.1-33B | National Institute of Informatics | 1680 ± 16 | Sep 28, 2026 | Open |
| 305 | GPT-OSS-Swallow-120B-RL-v0.1 | Institute of Science Tokyo | 1653 ± 20 | Feb 20, 2026 | Open |
| 332 | llm-jp-4-33B | National Institute of Informatics | 1640 ± 16 | Aug 18, 2026 | Open |
| 361 | llm-jp-4.1-32B-a3B | National Institute of Informatics | 1629 ± 15 | Sep 28, 2026 | Open |
| 369 | Qwen3-Swallow-32B-RL-v0.2 | Institute of Science Tokyo | 1627 ± 19 | Feb 19, 2026 | Open |
| 398 | llm-jp-4.1-8B | National Institute of Informatics | 1614 ± 15 | Sep 28, 2026 | Open |
| 439 | GPT-OSS-Swallow-20B-RL-v0.1 | Institute of Science Tokyo | 1601 ± 20 | Feb 20, 2026 | Open |
| 468 | llm-jp-4-32B-a3B-Thinking | National Institute of Informatics | 1593 ± 13 | Mar 16, 2026 | Open |
| 503 | ABEJA-Qwen2.5-32B-Japanese-v1.0 | ABEJA | 1582 ± 22 | May 14, 2025 | Open |
| 524 | llm-jp-4-8B-Thinking | National Institute of Informatics | 1576 ± 16 | Mar 16, 2026 | Open |
| 547 | ELYZA-Shortcut-1.0-Qwen-32B | ELYZA | 1569 ± 19 | Apr 30, 2025 | Open |
| 597 | Qwen3-Swallow-8B-RL-v0.2 | Institute of Science Tokyo | 1556 ± 18 | Feb 19, 2026 | Open |
| 688 | ELYZA-Thinking-1.0-Qwen-32B | ELYZA | 1528 ± 28 | May 1, 2025 | Open |
| 835 | llm-jp-3.1-13B-instruct4 | National Institute of Informatics | 1483 ± 26 | May 23, 2025 | Open |
| 874 | Llama-3.1-Swallow-8B-Instruct-v0.5 | Institute of Science Tokyo | 1473 ± 24 | Jun 25, 2025 | Open |
| 1273 | llm-jp-3.1-1.8B-instruct4 | National Institute of Informatics | 1385 ± 27 | May 27, 2025 | Open |
Interactive version: theaggregate.ai/country?slug=japan · How It Works · Data refreshed daily, snapshot 2026-10-20.