To ensure the best possible experience when using ScienceClaw, we strongly recommend selecting a base model with sufficient reasoning, tool-use, and long-context capabilities.
Based on our internal testing and real-world usage, we suggest using models at or above the capability level of:
• DeepSeek V3.2
• GLM 4.7+
• Kimi K2.5+
• GPT-5+
These models demonstrate strong performance in complex scientific workflows, multi-step reasoning, tool invocation, and long-horizon task execution — all of which are critical for fully leveraging ScienceClaw’s capabilities.
If you do not have access to a suitable base model, we recommend applying for 10M free DeepSeek V3.2 tokens via our Taichu Cloud platform.
⚠️Link：https://gateway.taichuai.cn/modelhub/login
Important Note:
Using weaker base models may significantly impact:
• Task completion quality
• Stability of multi-step workflows
• Tool-calling accuracy
• Overall user experience
If you encounter unexpected behavior or degraded performance, please first verify whether your selected model meets the recommended capability level.
To ensure the best possible experience when using ScienceClaw, we strongly recommend selecting a base model with sufficient reasoning, tool-use, and long-context capabilities.
Based on our internal testing and real-world usage, we suggest using models at or above the capability level of:
• DeepSeek V3.2
• GLM 4.7+
• Kimi K2.5+
• GPT-5+
These models demonstrate strong performance in complex scientific workflows, multi-step reasoning, tool invocation, and long-horizon task execution — all of which are critical for fully leveraging ScienceClaw’s capabilities.
If you do not have access to a suitable base model, we recommend applying for 10M free DeepSeek V3.2 tokens via our Taichu Cloud platform.
Important Note:
Using weaker base models may significantly impact:
• Task completion quality
• Stability of multi-step workflows
• Tool-calling accuracy
• Overall user experience
If you encounter unexpected behavior or degraded performance, please first verify whether your selected model meets the recommended capability level.