Models that expose an explicit `<think>` or chain-of-thought trace — DeepSeek-R1, Qwen-QwQ, GLM-Z1. Best for math-heavy, multi-step, or agentic workloads where the reasoning trace is itself valuable. Includes tool-use / function-calling tuned variants.
| Model | Creator | Context | Open weight | Tags |
|---|---|---|---|---|
| Yi Lightning | 01.AI | 16K | — | chat · reasoning |
| Qwen 2.5 Max | Alibaba Cloud | 33K | — | chat · reasoning · code |
| Qwen3 72B Instruct | Alibaba | 131K | Yes | chat · reasoning · multilingual |
| Doubao Pro 32K | ByteDance | 33K | — | chat · reasoning · code |
| DeepSeek Chat V3.1 | DeepSeek | 131K | Yes | chat · reasoning · code |
| DeepSeek R1 | DeepSeek AI | 66K | Yes | reasoning · math · coding |
| DeepSeek V3 | DeepSeek AI | 66K | Yes | reasoning · coding · tool_use |
| Llama 3.1 405B Instruct | Meta | 131K | Yes | chat · reasoning · code |
| Llama 3.3 70B Instruct | Meta | 131K | Yes | chat · coding · reasoning |
| GLM-4-Air | Zhipu AI | 131K | — | chat · reasoning · code |
Your brand here — Reach our audience of professional directory owners and boost your sales.