modelgrep

Open-Source Meta Models

Quick answer · Updated June 2026

The best open-weight Meta model is Llama 4 Maverick (18.4 intelligence) — downloadable from Hugging Face and self-hostable. Llama 3.3 70B Instruct (free) (14.5) and Llama 3.3 70B Instruct (14.5) round out the top three.

18.4Intelligence
72 t/sSpeed
$0.150Input /M
1.0MContext

The best open-weight AI models — downloadable from Hugging Face and self-hostable — ranked by intelligence. Open-source alternatives to proprietary frontier models.

  1. 1M
    llama-4-maverick
    ToolsJSONVision18.4 intel · $0.150/M · 72 t/s
    18.4
    Intelligence
  2. 2M
    llama-3.3-70b-instruct:free
    Tools14.5 intel · Free/M · 115 t/s
    14.5
    Intelligence
  3. 3M
    llama-3.3-70b-instruct
    ToolsJSON14.5 intel · $0.100/M · 115 t/s
    14.5
    Intelligence
  4. 4M
    llama-4-scout
    ToolsJSONVision13.5 intel · $0.100/M · 130 t/s
    13.5
    Intelligence
  5. 5M
    llama-3.1-70b-instruct
    ToolsJSON12.5 intel · $0.400/M · 28 t/s
    12.5
    Intelligence
  6. 6M
    llama-3.1-8b-instruct
    ToolsJSON11.8 intel · $0.020/M · 145 t/s
    11.8
    Intelligence
  7. 7M
    llama-3-70b-instruct
    JSON8.9 intel · $0.510/M · 18 t/s
    8.9
    Intelligence
  8. 8M
    llama-3.2-11b-vision-instruct
    JSONVision8.7 intel · $0.345/M · 35 t/s
    8.7
    Intelligence
  9. 9M
    llama-3-8b-instruct
    6.4 intel · $0.140/M · 63 t/s
    6.4
    Intelligence
  10. 10M
    llama-3.2-1b-instruct
    6.3 intel · $0.027/M · 169 t/s
    6.3
    Intelligence
  11. 11M
    llama-guard-4-12b
    JSONVision$0.180/M · 18 t/s · 120ms ttft
    Intelligence
  12. 12M
    llama-3.2-3b-instruct:free
    Free/M · 102 t/s · 223ms ttft
    Intelligence
  13. 13M
    llama-3.2-3b-instruct
    $0.051/M · 102 t/s · 223ms ttft
    Intelligence

Frequently asked

What is the best open-source Meta model?

The best open-weight Meta model is Llama 4 Maverick (18.4 intelligence) — downloadable from Hugging Face and self-hostable. Llama 3.3 70B Instruct (free) (14.5) and Llama 3.3 70B Instruct (14.5) round out the top three.

What's a good alternative to Llama 4 Maverick?

Llama 3.3 70B Instruct (free) (14.5) is the closest alternative on this metric, followed by Llama 3.3 70B Instruct (14.5). See the full ranking above for the tradeoffs.

How many Meta models are there?

modelgrep tracks 13 Meta models with live benchmarks, speed, latency and per-provider pricing, led on intelligence by Llama 4 Maverick. 13 of them qualify for this ranking.

More Meta rankings

All rankings