MiniMax Plans 2.7 Trillion Parameter Open-Source Model

:rotating_light: AI Tool Alert - July 8, 2026

MiniMax Plans 2.7 Trillion Parameter Open-Source Model

Chinese AI startup MiniMax is building a 2.7T parameter LLM (M3 Pro) and plans to open-source it in Q3 2026 — the largest open-weight model from China.

Key details:

• 2.7 trillion parameters: Would dwarf current open-source leaders like Llama 4 (~400B)


• Open-source planned: Full weights, fine-tuning freedom, no API dependency


• Q3 2026 timeline: Shipping within weeks, right as GPT-5.6 Sol goes public (July 9)


• Strategic move: Expands Chinese open-source AI footprint globally

Why it matters:

• Self-host at infrastructure cost only — eliminates per-token API fees


• Fine-tune for proprietary domains without OpenAI/Anthropic restrictions


• Supply chain resilience: high-capability alternative to US-only API providers

Use case: A dev company self-hosts M3 Pro, fine-tunes on their internal codebase and docs, runs a private AI assistant for code review — zero API costs, full data privacy.

⚠️ Caveat: Running 2.7T params needs ~100+ H100s. Actual benchmarks pending.

Link: https://www.reuters.com/world/asia-pacific/chinas-minimax-plans-launch-giant-27-trillion-parameter-model-2026-07-08/