Meta Superintelligence LabsLanguage modelSupersededOpen weights
Llama 4 Maverick
Meta's first natively multimodal MoE Llama; strong value, weaker reasoning.
Overall Intelligence
#54
of 56
29.2 pts behind #1
0
races led
0
weeks at #1
59.5
score
- Released
- Apr 5, 2025
- 1.5 yr ago
- Context
- 1M
- 16K max output
- Price
- $0.2 / $0.6
- per 1M in/out · $0.30 blended
- Speed
- 160 tok/s
- 0.55s to first token
- Modalities
- text · image
- outputs text
- Weights
- Open
- 400B MoE (17B active)
The 30-second version
Meta's spring-2025 open mixture-of-experts model. Popular for low latency, but it fell behind rivals quickly.
Key takeaways
- 1Open weights, natively multimodal.
- 2Benchmark controversy at launch dented trust.
- 3Succeeded by the Muse family in 2026.
Where it ranks
Current snapshot · illustrative scores · click any race to watch it
Llama 4Best on each axis
Overall
Coding
55Coding+35.6 pts53Frontend & UI+180 Elo55Backend & APIs+30.8 pts55Distributed Systems+34.4 pts56Full-Stack Apps+33.1 pts56Debugging & Repair+36.8 pts56Agentic Coding+38.8 pts56Code Review & Security+32.4 pts55Data Science & SQL+31.9 pts55Mobile Apps+31.3 pts55Terminal & DevOps+41.9 pts55Competitive Programming+1,160
Hard Reasoning
Math
Science
Writing & Language
Multimodal
Tool Use & Agents
Overall Intelligence over time
#1 frontierLlama 4V4.1 FlashGLM-5.3
Update history
The distilled version first; open any release for the deeper breakdown.