Meta AI

Llama 4 Maverick

Released Apr 5, 2025 · knowledge cutoff 2024-08

Status: Superseded
Location: United States
Type: LLM
Modality: Multimodal
Context window: 1M
Max output: 8K
Speed: —
Price ($/MTok in / out): $0.27 / $0.85
Cost per task: $0.029
Open weights: Yes

Benchmarks4 of 10 reported

MMLU-Proretired

80.5%

GPQA Diamond

69.8%

SWE-bench Pro

5.24%

LMArena Elo

1330

SWE-bench Verified

—

Terminal-Bench 2.1

—

AIME (latest)retired

—

Humanity's Last Exam

—

GDPval-AA v2

—

ARC-AGI-2

—

Strengths & weaknesses

Strengths

Sparse routing keeps per-token inference cheap for a model of its size
Accepts images natively, unlike the Llama 3 line
Broad hosting support from day one

Weaknesses

Day-to-day quality fell short of the launch impression, and trust never fully recovered
Weak at code and maths against open models released within months of it
Long-context quality degrades well before the advertised limit

Meta's last major open release before its superintelligence-lab reorg. SWE-bench Pro 5.24% is Scale's public-set leaderboard figure for llama4-maverick-17b-instruct — third-party, not Meta-reported.

For developers

API model strings

Not researched

Licence

Llama 4 Community License Agreement · Restricted · commercial use permitted

Retirement

No retirement announced

Lineage

No recorded predecessor

News