Minimax

MiniMax M-series production wiring patterns for the OpenAI-compatible API at api.minimax.io. Use when wiring MiniMax-M2.7 / MiniMax-M2.7-highspeed / Hailuo into a service (Karakeep tagging, Linkwarden summarization, RAG, agentic workflows, quant trade signals), debugging silent-dropped OpenAI parameters, configuring prompt caching (cache_control), handling rate limits via base_resp.status_code (no HTTP 429), choosing between plain M2.7 and -highspeed models, defending against the model's documented hallucination + saturation failure modes, or detecting model upgrades. Distilled from a 41-iteration M2.7-highspeed exploration campaign (40 verified pattern docs). TRIGGERS - MiniMax, MiniMax-M2.7, MiniMax-M2.7-highspeed, MiniMax-M2.5, MiniMax-M2, M-series reasoning model, api.minimax.io, OPENAI_BASE_URL minimax, OpenAI-compatible Chinese LLM provider, Chinese LLM API, Hailuo, hailuo.ai, MiniMax Hailuo, MiniMax embeddings, embo-01, MiniMax TTS, t2a_v2, speech-01-turbo, MiniMax video, video-01, video_generation, MiniMax tools, MiniMax function calling, MiniMax JSON mode, response_format silently dropped, stop parameter silently dropped, tool_choice silently dropped, image_url silently dropped, name field silently dropped, MiniMax cache_control, prompt_tokens_details cached_tokens, cache_creation_input_tokens, cache_read_input_tokens, base_resp status_code, status_code 1002, status_code 1004, status_code 2013, status_code 2061, RPM rate limit, MiniMax billing chatcompletion-v2, plus quant-LLM use cases — trade signal JSON, financial agentic stack, Tier F primitives, M2.7 saturation, M2.7 hallucination, Black-Scholes saturation, backtesting.py code generation hallucinated imports.

mkdir -p ~/.claude/skills/minimax
curl -L https://claudskills.com/skills/minimax/SKILL.md \
  -o ~/.claude/skills/minimax/SKILL.md