Swallow Models by Tokyo Institute of Technology
Last refreshed 2026-10-02. Next refresh: weekly.
Details
About
Swallow is the Tokyo Institute of Technology (tokyotech-llm) Japanese-focused LLM line built by continual pre-training of strong open bases, with instruction-tuned Instruct variants. Llama 3.1 Swallow (8B and 70B) continues Meta Llama 3.1 on a large Japanese web corpus plus Wikipedia, math, and code, then applies supervised fine-tuning on Japanese-oriented synthetic instruction data so Japanese ability rises while English capability is retained. The Swallow team publishes base and Instruct checkpoints across versions and documents the series on the Swallow project site and Hugging Face.
Decision facts
- Best fit
- codingmath-heavy prompts
- Capability starting point
- Swallow 13B Instruct with 8k context
- Lowest tracked input
- Not tracked
- Closest related family
- 360Zhinao
Current Variants
Use-when guidance is based on each model's tracked capabilities, context window, release date, and replacement status.
Use when the workload needs 4k context and 70B parameters.
Use when the workload needs 4k context and 8B parameters.
Use when the workload needs 8k context and 13B parameters.
Use when the workload needs 8k context and 13B parameters.
Use when the workload needs 4k context and 7B parameters.
Use when the workload needs 4k context and 7B parameters.
Use when the workload needs 4k context and 70B parameters.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Llama 3.1 Swallow 70B Instruct | Use when the workload needs 4k context and 70B parameters. | 2025-01 | 4k context70B parameters | Current |
| Llama 3.1 Swallow 8B Instruct | Use when the workload needs 4k context and 8B parameters. | 2025-01 | 4k context8B parameters | Current |
| Swallow 13B Instruct | Use when the workload needs 8k context and 13B parameters. | 2024-12 | 8k context13B parameters | Current |
| Swallow 13B | Use when the workload needs 8k context and 13B parameters. | 2024-12 | 8k context13B parameters | Current |
| Swallow 7B Instruct | Use when the workload needs 4k context and 7B parameters. | 2024-09 | 4k context7B parameters | Current |
| Swallow 7B | Use when the workload needs 4k context and 7B parameters. | 2024-09 | 4k context7B parameters | Current |
| Llama 3 Swallow 70B Instruct | Use when the workload needs 4k context and 70B parameters. | 2024-06 | 4k context70B parameters | Current |
Release Timeline
4 release groupsSpecifications(7 models)
| Model | Released | Context | Parameters |
|---|---|---|---|
| Llama 3.1 Swallow 70B Instruct | 2025-01 | 4k | 70B |
| Llama 3.1 Swallow 8B Instruct | 2025-01 | 4k | 8B |
| Swallow 13B Instruct | 2024-12 | 8k | 13B |
| Swallow 13B | 2024-12 | 8k | 13B |
| Swallow 7B Instruct | 2024-09 | 4k | 7B |
| Swallow 7B | 2024-09 | 4k | 7B |
| Llama 3 Swallow 70B Instruct | 2024-06 | 4k | 70B |
Available From(1 provider)
Popular comparisons in this family
- Llama 3.1 Swallow 8B Instruct vs Qwen3.5-9B318
- ELYZA Japanese Llama 2 13B vs Llama 3.1 Swallow 8B Instruct120
- ELYZA Japanese Llama 2 7B vs Llama 3 Swallow 70B Instruct103
- ELYZA Japanese Llama 2 7B vs Llama 3.1 Swallow 8B Instruct97
- Llama 3.1 Swallow 8B Instruct vs Qwen2-7B-Instruct87
- Gemma 3 12B vs Llama 3.1 Swallow 8B Instruct84
- ELYZA Japanese Llama 2 13B vs Llama 3 Swallow 70B Instruct77
- Llama 3.1 Swallow 8B Instruct vs Qwen3.5-4B65
- Bielik 11B v2.6 Instruct vs Llama 3 Swallow 70B Instruct36
- GPT-4 vs Llama 3.1 Swallow 8B Instruct33
Models(7)
Llama 3.1 Swallow 70B Instruct
Llama 3.1 Swallow 8B Instruct
Swallow 13B Instruct
Swallow 13B
Swallow 7B Instruct
Swallow 7B
Llama 3 Swallow 70B Instruct