Swallow - Post-trained Japanese - JamC-QA: leaderboard
From the Swallow LLM Leaderboard (Institute of Science Tokyo) post-trained Japanese board, which evaluates instruction-tuned and reasoning models with the Swallow evaluation harness. JamC-QA (Oka et al., 2025) is multiple-choice question answering on Japan-specific knowledge, run zero-shot with a chain-of-thought prompt; metric is accuracy. Upstream publishes 0-1 scores; shown as 0-100, higher is better.
Metric: Accuracy (%). Source: swallow-llm.github.io. Saturation forecast: Around February 2027. 68 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.4 (2026-03-05) (Medium) | 89.1 | #63 (GPT-5.4 (2026-03-05)) |
| 2 | GPT-5 (Medium) | 85.8 | #105 (GPT-5) |
| 3 | O3 (2025-04-16) (Medium) | 85.1 | #125 (O3 (2025-04-16)) |
| 4 | GPT-4.1 | 79 | #258 |
| 5 | GPT-4o (2024-08-06) | 74.7 | #343 |
| 6 | GPT-5 Mini (2025-08-07) (Medium) | 70.1 | #175 (GPT-5 Mini (2025-08-07)) |
| 7 | Gemma 4 31B (IT) (Thinking) | 68.6 | #178 (Gemma 4 31B (IT)) |
| 8 | Qwen 3.5 122B A10B (Thinking) | 67.1 | #184 (Qwen 3.5 122B A10B) |
| 9 | gemma-4-26B-A4B-it (Thinking) | 64.7 | |
| 10 | Qwen 3 235B A22B 2507 (Thinking) | 64.1 | #263 |
| 11 | Qwen 3 235B A22B 2507 Instruct | 63.6 | #315 |
| 12 | GPT-OSS-Swallow-120B-RL-v0.1 (Thinking) | 61.6 | #309 (GPT-OSS-Swallow-120B-RL-v0.1) |
| 13 | Qwen 3 Next 80B A3B (Thinking) | 61.4 | #312 |
| 14 | Qwen 3.5 35B A3B (Thinking) | 61.1 | #268 (Qwen 3.5 35B A3B) |
| 15 | Qwen 3 Next 80B A3B Instruct | 59.9 | #381 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=swallow-post-trained-japanese-jamc-qa · How It Works · Data refreshed daily, snapshot 2026-11-05.