Meta: Llama 4 Maverick, Meta — route, pricing & source
Meta open-weight multimodal MoE model with 17B active and 400B total parameters.
This model does not have reliable token pricing or may use unit-based billing. Example costs on this page should not be used as an official or real invoice estimate for this model.
Meta does not sell a direct Llama API price. API prices depend on the hosting provider; weights can be self-hosted under the Llama 4 Community License.
Context and limits
From source metadata + verification fields — confirm provider docs before production.
- MMLU Proreasoning_knowledge · provider_reported80.5%Provider
Instruction-tuned model, provider-reported model card result.
- GPQA Diamondreasoning_knowledge · provider_reported69.8%Provider
Instruction-tuned model, provider-reported model card result.
- LiveCodeBenchcoding · provider_reported43.4%Provider
Instruction-tuned model, provider-reported model card result.
- LMArena Vision Arenaoverall · official_benchmark_api1,147Independent
Rank: 80 | Votes: 7063 | 95% CI: 1138.3-1156.6 | Published: 2026-05-18 | Model license label in dataset: Llama 4
Access and license
Example costs
Scenario: 1K input + 300 output tokens / request
Cheaper alternatives
Lower-cost models for the same scenario
No cheaper alternative for this scenario.
Source
- This is an open-weight model profile, not a direct hosted API price.
- Benchmark values are from Meta's official Hugging Face model card and should be labeled as provider-reported.
- Includes Hugging Face benchmark data only when the dataset is public, ungated, allowlisted, and matched by reviewed aliases.
