MiniMax H3 local was evaluated in the Pelican test, using a Qwen3 text encoder for prompting and achieving a 10‑second clip with 91‑second generation time on an RTX 5090, while its status as a traditional LLM is pending moderator decision.