Two models together

[frkr]

Description

I want to use Claude 4.5 Sonnet or GPT-5.1 as architecture, documentation, Mermaid.js, and arbitration. And use Qwen or Grok for implementation.

Can both sides work together to create a leader and a worker?

[BrunoKrugel]

You can change agent without starting a new session, just ask Claude to plan the architecture, documentation, Mermaid.js, and arbitration.

Then change the agent and ask to implement it.

[frkr]

i know that @BrunoKrugel, but this its not the point... i can use, for example, a local LLM for toolcalling and MCP and Claude for design the archtecture and Grok or Perplexity for researching solution. Or a local LLM for turnovers, tasks definitions, drilldown, task tracking on an a document... and QWEN for coding.

The point is, on the same history and the same call. Not only on the same history.

[meowgorithm]

So basically the ability to choose models on the task-level. It's a clever idea to be sure: converting to a discussion in order to work this out further and keep the issue tracker clean. If we reach a consensus we'll convert back to an issue.

[frkr]

I've been thinking about it these days. I believe this fulfills my request.

If I can choose the model by: Agent, sub-agent, skill, mode and task. It will work out.

Because I can use gemini flash to control the context of the chat and the tasks and skills will be executed by other models. This is going to be brilliant.

[frkr]

Here are the scenario:

Making an spec with Perplexity (for best pratices and macro guidelines), converting to an elaborating the nexts step with AGENTS.md and 'TASKS.md' with GPT-5-Codex-Max. When plan its finished, code with "o3-mini or GPT-4.1o" thinking but "gpt-5-codex-mini" or "claude-haiku" only for coding, to write!

1. Plan, GPT primary and Perplexity as subagent.
2. Leading, GPT-4.1o
3. Coding, GPT-5-Codex-Mini

or

1. Plan, with Opus and Gemini as subagent
2. Leading, Local model on Ollama
3. Coding, Grok Code

[frkr]

and... whats is:

"leading": Updating checklist TODO using TASKS.md, making code review if that on plan and AGENTS.md are implemented or need a drilldown.