| OpenAI Chat | reasoning_effort | none/minimal/low/medium/high/xhigh/max(按模型) | gpt-6.1-sol 无 none;gpt-6-luna/6.1-terra 支持 none |
| OpenAI Responses | reasoning.effort + reasoning.summary | 同上;summary auto/concise/detailed | 非 none 时不要传 temperature/top_p |
| Claude 原生 | thinking:{type:adaptive} + output_config.effort | low/medium/high/xhigh/max | Claude 5.x 不再接受 type:enabled + budget_tokens |
| Gemini 原生 | generationConfig.thinkingConfig.thinkingLevel | low/medium/high | 2.5 系列用 thinkingBudget |
| DeepSeek V4 | thinking:{type} + reasoning_effort | enabled/disabled;low/high/max | |
| GLM-5.3 | thinking:{type:enabled} + reasoning_effort | low/high/max(默认 max) | 不能 disabled |
| New API 后缀 | 模型名后加 -high / -low / -none 等 | OpenAI GPT 系列;Claude/Gemini 需站点开启思考适配 | Claude/Gemini 还支持 -thinking、-nothinking |
| New API 修饰符 | 模型名后加 @thinking:on|off|adaptive|数字 | 任意渠道 | 例:claude-sonnet-5-5@thinking:adaptive |