Summary

This PR adds MiniMax as a supported LLM provider for all Agent-S versions (s2, s2_5, s3), with MiniMax-M3 as the default model.

• Adds LMMEngineMiniMax engine class using the Anthropic-compatible API interface (https://api.minimax.io/anthropic)
• Supported models: MiniMax-M3 (default, 512K context, 128K max output, image input), MiniMax-M2.7, MiniMax-M2.7-highspeed
• Handles MiniMax-specific constraints: temperature must be in (0.0, 1.0] (clamps zero to 1.0), filters unsupported parameters (top_k, stop_sequences, service_tier, etc.)
• Registers engine_type: "minimax" in all three LMMAgent classes
• Updates models.md with MiniMax setup instructions
• Adds 20 unit tests covering instantiation, temperature handling, parameter filtering, API key resolution, model name pass-through (M3 + M2.7-highspeed), and LMMAgent integration

Why MiniMax-M3 as default

MiniMax-M3 is the latest model in the MiniMax family:

• 512K context window (up from 192K on M2.7)
• 128K max output
• Image input support on both OpenAI-compatible and Anthropic-compatible endpoints
• Pricing (≤512K tier): $0.6/M input, $2.4/M output, $0.12/M cache reads

M2.7 and M2.7-highspeed are kept as alternatives for users who prefer them.

Usage

from gui_agents.s2_5.core.mllm import LMMAgent

engine_params = {
    "engine_type": "minimax",
    "model": "MiniMax-M3",  # default; also "MiniMax-M2.7" / "MiniMax-M2.7-highspeed"
}
agent = LMMAgent(engine_params=engine_params)

Set MINIMAX_API_KEY environment variable (optionally MINIMAX_BASE_URL).

API References

• Chat (Anthropic Compatible): https://platform.minimax.io/docs/api-reference/text-anthropic-api