Allow more agent model selection
As we all know AI is moving really fast and there's new models every few weeks. In order to get the best out of bao, I think it would be good if:
- The model selectors are updated to use the latest models (Gemini 3.8 instead of 3.7, GPT-6 instead of 5.6, allowing Claude Opus 5.5)
- We get the ability to choose completely custom models. Might not be the best usability-wise for now but could also be hidden in an "Advanced" mode if you want to keep the defaults simpler for users and curated models you know that work
6 Comments
Sign in to comment
I’d add that it would be good to be able to have multiple models attached; something like a user-specified default and an option to pick the model for a specific task
That would help my use case of local LLMs/SLMs as well. I have a machine with GPU dedicated for this kind of task
I haven't integrated Anytwo with my local LLM API it yet, but it seems in order to work with Anytwo I would need to choose some vendor API format and create aliases using the closed models names (claude, gpt) for open-weight models (gemma, qwen, etc)
I see the need, but my provision would be that we're not loading the team with a never-ending task updating the list for the host of new models that are appearing. I don't know enough about how this works behind the scenes to wonder sensibly whether they could offload to OpenClaw or Opentouter?
100% agree. Maybe the team can keep builtin model sets only for the main providers, targeting the needs of most users, and leave the rest to the "Advanced mode" (I like "Custom mode" terminology more), where we can choose:
- API spec/style (OpenAI, Anthropic, LM Studio, etc)
- endpoint URL
- Bearer token (optional)
- tier X model ID mapping
Cross-referencing related post:
Thanks - I’d seen that post and promptly forgotten it :scream: