N
NexusRoute
Back to Models

Llama 4 Maverick

Metafrontier

Meta's most capable open-weight model. A 400B MoE (17B active) with native multimodal support and a 1M token context window. Approaches closed-source frontier quality on many benchmarks while being fully open-weight. Competitive with GPT-5.4-mini and Claude Sonnet 4.6.

Released 2025-11-14Knowledge cutoff: 2025-08
Medium confidence|Updated 72d ago|88% source confidence

Specifications

Context Window

1M tokens

Max Output

64K tokens

Input Price

$0.500 / 1M tokens

Output Price

$1.50 / 1M tokens

Latency Tier

Moderate (speed score: 6.5/10)

Capability Profile

Long Context
9/10
Reasoning
8/10
Coding
8/10
Cost Efficiency
8/10
Factuality
8/10
Structured Output
7.5/10
Multimodal
7.5/10
Creativity
7.5/10
Instruction Following
7.5/10
Tool Use
7.5/10
Conversational
7.5/10
Speed
6.5/10
Safety & Enterprise
6.5/10

Feature Support

Vision Yes
Audio In No
Audio Out No
Video No
Image Generation No
Image Editing No
Function Calling Yes
JSON Mode Yes
Structured Output Yes
Streaming Yes
Reasoning No
Realtime No
Computer Use No
Web Search No

Best Use Cases

Open-weight deployments needing frontier-class quality
Fine-tuning for enterprise or domain-specific applications
Long-context multimodal analysis at open-weight pricing
Research requiring model weight access for interpretability or customization
Cost-effective frontier alternative to GPT-5.4-mini or Sonnet 4.6

Not Ideal For

Simple self-hosted inference on a single GPU (use Scout instead)
Enterprise deployments requiring the strictest safety alignment
Audio or video processing
Applications where the absolute best coding quality is needed (use Claude Opus)

Strengths

Best open-weight model available — approaches closed-source frontier quality
1M context with strong recall — best long-context open model
Native multimodal (text + images) with strong vision understanding
Full model weights available for customization and fine-tuning
Excellent value through hosted providers at ~$0.50/$1.50

Weaknesses

Requires multi-GPU infrastructure for self-hosting (4-8 H100s recommended)
Quality gap vs Claude Opus 4.6 and GPT-5.4 is measurable on hard tasks
Instruction following less precise than Anthropic or OpenAI models
Safety alignment is basic compared to closed-source competitors
Structured output compliance is good but not best-in-class

Edge Cases & Notes

17B active parameters per token despite 400B total — MoE efficiency is key
Pricing varies significantly by hosted provider
Fine-tuned variants from the community can significantly improve domain-specific performance
Self-hosting requires expertise in distributed inference (vLLM, TGI, etc.)

Provider Notes

Open-weight under Meta's Llama license. Available through Together AI, Fireworks, Replicate, and self-hosted. Self-hosting requires multi-GPU setup. The best open-weight option for teams needing frontier-class quality with weight access.

Benchmarks

MMLU88.5%
HumanEval88%
Arena Elo1340

Benchmark Notes

MMLU-Pro 88.5%. Competitive with GPT-5.4-mini on many benchmarks. SWE-bench ~48%. Best open-weight model on LMSYS Arena. Long-context performance is strong throughout 1M window.

Research Meta

Last Evaluated

2026-03-15

Source Confidence

88%

Evaluation Method

Open LLM Leaderboard, LMSYS Arena, SWE-bench, long-context evaluation

Needs Re-evaluation

No

Sources

  • Meta Llama 4 technical report
  • Open LLM Leaderboard
  • LMSYS Chatbot Arena
  • SWE-bench leaderboard

Continue exploring

Route a prompt

See how Llama 4 Maverick ranks

Compare models

Side-by-side analysis

Browse registry

Explore all 24 models