最佳聊天AI模型
按对话能力排名的语言模型。 更新于2026年8月。基于独立基准数据。
| # | 模型 | 提供商 | 分数 | 速度 | 价格（输出） |
|---|---|---|---|---|---|
| 1 | GLM-5.3 Flash | Z.ai | 80.1 | 182 t/s | $0.3/M |
| 2 | Qwen3.8 Flash Next | Alibaba | 79.6 | 168 t/s | $0.4/M |
| 3 | Gemini 3.1 Pro | 76.8 | 136 t/s | $12/M | |
| 4 | HyperNova 60B | Multiverse | 75.1 | 367 t/s | $0.14/M |
| 5 | Gemini 3.8 Flash | 73.3 | 302 t/s | $3.75/M | |
| 6 | Gemini 3.8 Flash (Medium Reasoning) | 73.3 | 315 t/s | $3.75/M | |
| 7 | Gemini 3.8 Flash (Low Reasoning) | 73.3 | 340 t/s | $3.75/M | |
| 8 | GPT-6 Sol (Non-reasoning) | OpenAI | 72.9 | 265 t/s | $4/M |
| 9 | GPT-6 Sol (xhigh) | OpenAI | 70.5 | 210 t/s | $4/M |
| 10 | GPT-6 Sol | OpenAI | 70.5 | 220 t/s | $4/M |
| 11 | GPT-6 Sol (high) | OpenAI | 70.5 | 226 t/s | $4/M |
| 12 | GPT-6 Sol (medium) | OpenAI | 70.5 | 238 t/s | $4/M |
| 13 | GPT-6 Sol (low) | OpenAI | 70.5 | 250 t/s | $4/M |
| 14 | Granite 4.2 3B Instruct | IBM | 69.6 | 148 t/s | $0.2/M |
| 15 | Qwen3.7 Max | Alibaba | 69.1 | 203 t/s | $7.5/M |
| 16 | LFM2.5-8B-A1B | Liquid AI | 68.1 | 338 t/s | $2/M |
| 17 | LFM2.5-VL-1.6B | Liquid AI | 68.1 | 395 t/s | $2/M |
| 18 | DeepSeek V4.1 Flash (Non-reasoning) | DeepSeek | 66.9 | 240 t/s | $1.2/M |
| 19 | DeepSeek V4.1 Flash | DeepSeek | 66.4 | 214 t/s | $1.2/M |
| 20 | DeepSeek V4 Flash 0420 (Non-reasoning) | DeepSeek | 66.0 | 210 t/s | $1.1/M |
GLM-5.3 Flash
Qwen3.8 Flash Next
Gemini 3.1 Pro
GPT-6 Sol (Non-reasoning)
Granite 4.2 3B Instruct
LFM2.5-8B-A1B
DeepSeek V4.1 Flash (Non-reasoning)