Local LLM model page

MiroThinker v1.5 (30B MoE)

⚠️ Despite the small active count, this is a full 30B MoE model (Qwen3-30B-A3B base). ~82 GB full weights (Q4_K_M ≈18 GB). Deep-research agent with 256K context, tool calls, multilingual (EN/ZH). Requires H100 80 GB or serious multi-GPU. Not suitable for M1/M2 or consumer GPUs. Apache 2.0.

Find the best model for my hardwareBrowse all 183 LLMs
Parameters
30B (3B active, MoE)
Minimum RAM
48 GB
Model size
18 GB
Quantization
Q4_K_M

Can MiroThinker v1.5 (30B MoE) run locally?

MiroThinker v1.5 (30B MoE) is best suited for high-end workstations with 64 GB RAM. LocalClaw recommends Q4_K_M as the default quantization, with at least 48 GB RAM.

Search term for LM Studio or compatible runtimes: mirothinker-v1.5-30b

Hugging Face repository: miromind-ai/MiroThinker-v1.5-30B

reasoningcodepowerquality

Strengths

  • ⚠️ Despite the small active count, this is a full 30B MoE model (Qwen3-30B-A3B base). ~82 GB full weights (Q4_K_M ≈18 GB). Deep-research agent with 256K context, tool calls, multilingual (EN/ZH). Requires H100 80 GB or serious multi-GPU. Not suitable for M1/M2 or consumer GPUs. Apache 2.0.

Limitations

  • Performance depends heavily on quantization, RAM bandwidth and runtime support.

Best use cases

  • reasoning
  • code
  • power
  • quality

Benchmarks

Speed: 3/10

Quality: 9/10

Coding: 8/10

Reasoning: 10/10

Technical details

Developer: mirothinker

License: See model repository

Context window: Unknown tokens

Architecture: See model card

Released: 2026-02