Best local LLMs for the Rockchip RK3568 NAS (4 GB)

Each list assumes the context that task needs, so a model that fits for chat may rank lower for long documents.

General chat at 8K context

  1. 47Fair quality (C) · 47 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 69/100 (quality, fit, speed and memory use) sets the ranking. LFM3 1.7B Q3_K_M · 3.3 tok/s · Slow (tight fit)
  2. 35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. LFM3-VL 2B Q2_K · 3.7 tok/s · Slow
  3. 34Basic quality (D) · 34 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 65/100 (quality, fit, speed and memory use) sets the ranking. MiniCPM5 1B Q4_K_M · 4.0 tok/s · Slow
  4. 33Basic quality (D) · 33 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. MiniCPM-V 4.7 1B Q3_K_M · 3.6 tok/s · Slow
  5. 30Basic quality (D) · 30 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. LFM2.5 1.2B Instruct Q4_K_M · 3.7 tok/s · Slow

Coding at 32K context

  1. 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. PaddleOCR-VL 1.6 0.9B Q4_K_M · 4.5 tok/s · Slow (tight fit)
  2. 39Basic quality (D) · 39 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. LFM3 1.7B Q2_K · 4.1 tok/s · Slow (tight fit)
  3. 25Basic quality (D) · 25 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. MiniCPM5 1B Q2_K · 6.6 tok/s · Slow (tight fit)
  4. 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. MiniCPM-OCR 1.0 1.3B Q4_K_M · 3.4 tok/s · Slow (tight fit)
  5. 35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. LFM3-VL 2B Q2_K · 3.7 tok/s · Slow (tight fit)

Reasoning at 32K context

  1. 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. PaddleOCR-VL 1.6 0.9B Q4_K_M · 4.5 tok/s · Slow (tight fit)
  2. 39Basic quality (D) · 39 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. LFM3 1.7B Q2_K · 4.1 tok/s · Slow (tight fit)
  3. 25Basic quality (D) · 25 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. MiniCPM5 1B Q2_K · 6.6 tok/s · Slow (tight fit)
  4. 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. MiniCPM-OCR 1.0 1.3B Q4_K_M · 3.4 tok/s · Slow (tight fit)
  5. 35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. LFM3-VL 2B Q2_K · 3.7 tok/s · Slow (tight fit)

Documents (RAG) at 32K context

  1. 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 67/100 (quality, fit, speed and memory use) sets the ranking. PaddleOCR-VL 1.6 0.9B Q4_K_M · 4.5 tok/s · Slow (tight fit)
  2. 39Basic quality (D) · 39 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. LFM3 1.7B Q2_K · 4.1 tok/s · Slow (tight fit)
  3. 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 66/100 (quality, fit, speed and memory use) sets the ranking. MiniCPM-OCR 1.0 1.3B Q4_K_M · 3.4 tok/s · Slow (tight fit)
  4. 35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 64/100 (quality, fit, speed and memory use) sets the ranking. LFM3-VL 2B Q2_K · 3.7 tok/s · Slow (tight fit)
  5. 33Basic quality (D) · 33 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 63/100 (quality, fit, speed and memory use) sets the ranking. MiniCPM-V 4.7 1B Q3_K_M · 3.6 tok/s · Slow (tight fit)

Agents & tools at 64K context

  1. 7Basic quality (D) · 7 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. LFM2.5 230M Q2_K · 31.2 tok/s · Runs great
  2. 2Basic quality (D) · 2 of 100: how capable the model is at Q5_K_M, whatever it runs on. Match score 56/100 (quality, fit, speed and memory use) sets the ranking. Falcon-H1 Tiny 90M Instruct Q5_K_M · 40.1 tok/s · Runs great
  3. 21Basic quality (D) · 21 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. Qwen3.5 0.8B Q3_K_M · 6.5 tok/s · Slow (tight fit)
  4. 9Basic quality (D) · 9 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 54/100 (quality, fit, speed and memory use) sets the ranking. LFM2.5 350M Q2_K · 20.5 tok/s · Runs well (tight fit)
  5. 8Basic quality (D) · 8 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 53/100 (quality, fit, speed and memory use) sets the ranking. LFM2 350M Q2_K · 20.5 tok/s · Runs well (tight fit)

Reverse engineering at 32K context

  1. 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 62/100 (quality, fit, speed and memory use) sets the ranking. PaddleOCR-VL 1.6 0.9B Q4_K_M · 4.5 tok/s · Slow (tight fit)
  2. 39Basic quality (D) · 39 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. LFM3 1.7B Q2_K · 4.1 tok/s · Slow (tight fit)
  3. 39Basic quality (D) · 39 of 100: how capable the model is at Q4_K_M, whatever it runs on. Match score 61/100 (quality, fit, speed and memory use) sets the ranking. MiniCPM-OCR 1.0 1.3B Q4_K_M · 3.4 tok/s · Slow (tight fit)
  4. 35Basic quality (D) · 35 of 100: how capable the model is at Q2_K, whatever it runs on. Match score 59/100 (quality, fit, speed and memory use) sets the ranking. LFM3-VL 2B Q2_K · 3.7 tok/s · Slow (tight fit)
  5. 33Basic quality (D) · 33 of 100: how capable the model is at Q3_K_M, whatever it runs on. Match score 58/100 (quality, fit, speed and memory use) sets the ranking. MiniCPM-V 4.7 1B Q3_K_M · 3.6 tok/s · Slow (tight fit)

Estimates come from published hardware specs and our open formulas, not from running the model on your machine. Real speed varies with runtime, drivers and settings. Numbers marked “Measured” come from a linked benchmark.

Was this page helpful?