Qwen2 Models by Alibaba
Last refreshed 2026-10-17. Next refresh: weekly.
Details
Capabilities
About
Qwen2 is Alibaba’s Qwen Team language-model generation that followed Qwen1.5, released as base and instruction-tuned sizes from Qwen2-0.5B through Qwen2-72B, including the MoE Qwen2-57B-A14B. The series emphasizes Grouped-Query Attention across sizes, stronger coding and math than the prior generation, broader multilingual training beyond English and Chinese, and long-context instruction models (up to 128K with extrapolation aids on the larger Instruct variants). Checkpoints were published on Hugging Face and ModelScope.
Decision facts
- Best fit
- structured outputscodingmath-heavy prompts
- Capability starting point
- Qwen2-0.5B
- Lowest tracked input
- Qwen2-7B · $0.050/1M · DeepInfra
- Closest related family
- Ember
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 128K context and 7B parameters.
Use when the workload needs 32K context, 72B parameters, and structured outputs.
Use when the workload needs 32K context, 7B parameters, and structured outputs.
Use when the workload needs 57.4B parameters and structured outputs.
| Model | Use when | Released | Signals | Status |
|---|---|---|---|---|
| Qwen2-7B-Instruct | Use when the workload needs 128K context and 7B parameters. | 2024-06 | 128K context7B parameters | Current |
| Together AI Qwen2-72B-Instruct | Use when the workload needs 32K context, 72B parameters, and structured outputs. | 2024-06 | 32K context72B parametersstructured outputs | Current |
| Together AI Qwen2-7B-Instruct | Use when the workload needs 32K context, 7B parameters, and structured outputs. | 2024-06 | 32K context7B parametersstructured outputs | Current |
| Qwen2-0.5B | Use when the workload needs 490M parameters. | 2024-06 | 490M parameters | Current |
| Qwen2-57B-A14B | Use when the workload needs 57.4B parameters and structured outputs. | 2024-06 | 57.4B parametersstructured outputs | Current |
Release Timeline
1 release groupReplaced By
Keep for legacy integrations; evaluate Qwen2.5-1.5B before new work.
Keep for legacy integrations; evaluate Qwen2.5-72B before new work.
Keep for legacy integrations; evaluate Qwen2.5-7B before new work.
Specifications(8 models)
| Model | Released | Context | Parameters | Structured Outputs |
|---|---|---|---|---|
| Qwen2-7B-Instruct | 2024-06 | 128K | 7B | No |
| Together AI Qwen2-72B-Instruct | 2024-06 | 32K | 72B | Yes |
| Together AI Qwen2-7B-Instruct | 2024-06 | 32K | 7B | Yes |
| Qwen2-0.5B | 2024-06 | — | 490M | No |
| Qwen2-57B-A14B | 2024-06 | — | 57.41B | Yes |
Available From(6 providers)
Pricing
| Model | Provider | Input / 1M | Output / 1M | Type |
|---|---|---|---|---|
| Together AI Qwen2-7B-Instruct | Together AI | $0.15 | $0.15 | Serverless |
| Qwen2-57B-A14B | DeepInfra | $0.16 | $0.16 | Serverless |
| Together AI Qwen2-72B-Instruct | Together AI | $0.7 | $0.7 | Serverless |
Popular comparisons in this family
- Llama 3.1 Swallow 8B Instruct vs Qwen2-7B-Instruct87
- GPT-5.4 vs Together AI Qwen2-72B-Instruct84
- MiniCPM-V 4.6 vs Qwen2-7B-Instruct76
- Kimi K2.5 vs Together AI Qwen2-72B-Instruct73
- ELYZA Japanese Llama 2 7B vs Qwen2-7B-Instruct67
- Llama 2 7B Chat vs Qwen2-7B-Instruct64
- Gemini 2.5 Flash vs Qwen2-7B-Instruct64
- GLM-5.1 vs Together AI Qwen2-7B-Instruct54
- DeepSeek V4 Flash vs Together AI Qwen2-7B-Instruct52
- Llama 2 7B vs Qwen2-7B-Instruct51
