{"slug":"llama-4-maverick","id":"llama-4-maverick","type":"model","title":"Llama 4 Maverick","description":"Meta's flagship open-source MoE model with 400B total parameters (17B active). Scored #2 on LMArena leaderboard (ELO 1,417). Native multimodal capabilities.","last_updated":"2026-04-10","last_verified":null,"verification_status":"unverified","markdown_url":"/content/models/llama-4-maverick.md","html_url":"/models/llama-4-maverick","api_url":"/api/v1/models/llama-4-maverick.json","content_hash":"676662087b7879291e437f5765c1cb309db3bb0227f69c8e23f32f3e584a5ba1","sha256":"676662087b7879291e437f5765c1cb309db3bb0227f69c8e23f32f3e584a5ba1","provider":"Meta","pricing":{"input":"Free (self-hosted)","output":"Free (self-hosted)","free":true,"note":"Or use via third-party APIs"},"benchmarks":{"reasoning":87,"coding":82,"math":84,"writing":86,"multilingual":84,"speed":80},"tags":["meta","open-source","text","image"],"website":"https://llama.meta.com","release_date":"2025-04","relationships":{"links":[],"related":[{"id":"llama-4-scout","title":"Llama 4 Scout","type":"model","html_url":"/models/llama-4-scout","markdown_url":"/content/models/llama-4-scout.md","shared_tags":["meta","open-source","text","image"],"score":8},{"id":"gemma-3","title":"Gemma 3","type":"model","html_url":"/models/gemma-3","markdown_url":"/content/models/gemma-3.md","shared_tags":["open-source","text","image"],"score":5},{"id":"gemma-4","title":"Gemma 4","type":"model","html_url":"/models/gemma-4","markdown_url":"/content/models/gemma-4.md","shared_tags":["open-source","text","image"],"score":5},{"id":"mistral-small-4","title":"Mistral Small 4","type":"model","html_url":"/models/mistral-small-4","markdown_url":"/content/models/mistral-small-4.md","shared_tags":["open-source","text","image"],"score":5},{"id":"qwen-3.5","title":"Qwen 3.5 397B-A17B","type":"model","html_url":"/models/qwen-3.5","markdown_url":"/content/models/qwen-3.5.md","shared_tags":["open-source","text","image"],"score":5},{"id":"claude-haiku-4.5","title":"Claude Haiku 4.5","type":"model","html_url":"/models/claude-haiku-4.5","markdown_url":"/content/models/claude-haiku-4.5.md","shared_tags":["text","image"],"score":4}],"explicit":{}},"metadata":{"title":"Llama 4 Maverick","type":"model","id":"llama-4-maverick","provider":"Meta","model_type":"open-source","release_date":"2025-04","description":"Meta's flagship open-source MoE model with 400B total parameters (17B active). Scored #2 on LMArena leaderboard (ELO 1,417). Native multimodal capabilities.","last_updated":"2026-04-10","context_window":"1M tokens","website":"https://llama.meta.com","license":"Llama Community License","modality":["text","image"],"tags":["meta","open-source","text","image"],"pricing":{"input":"Free (self-hosted)","output":"Free (self-hosted)","free":true,"note":"Or use via third-party APIs"},"benchmarks":{"reasoning":87,"coding":82,"math":84,"writing":86,"multilingual":84,"speed":80},"parameters":"400B total (17B active)","hardware_requirements":"4x A100 80GB or 8x RTX 4090 (FP16); single GPU with heavy quantization","best_for":["Self-hosted applications","Fine-tuning","Privacy-sensitive tasks","Production deployment"]},"content_text":"# Llama 4 Maverick\n\nThe open-source model that got closest to proprietary quality. Maverick hit #2 on the LMArena leaderboard (ELO 1,417) and its MoE architecture means only 17B of its 400B parameters are active per inference -- making it surprisingly efficient to run for something this capable.\n\nFor self-hosting, Maverick is the safe default. The Llama Community License is permissive enough for most commercial use, the community is massive, and the fine-tuning ecosystem is the most mature in open source. If you need a model you control completely -- for privacy, compliance, or cost predictability -- Maverick should be your first evaluation.\n\nThe honest gap: coding. At 40% on LiveCodeBench versus 85% for GPT-5, Maverick is not competitive for serious software engineering work. The 82/100 coding score looks decent in isolation but lags behind DeepSeek V3.2 (88), Qwen 3 (90), and every frontier proprietary model. Writing (86) and reasoning (87) are solid but not exceptional. You need 4x A100 80GB GPUs for full-precision hosting, though quantized versions can run on consumer hardware.\n\n**When to pick something else:** For coding, DeepSeek V3.2 or Qwen 3 are better open-source options. For maximum quality without self-hosting constraints, Claude Sonnet 4.6 at $3/$15 will outperform Maverick on every benchmark. But if you need the largest open-source community, the deepest integration ecosystem, and a model you can fine-tune and deploy on your own infrastructure, Maverick remains the standard choice.","content_length":2494,"generated_at":"2026-04-24"}