Llama 4 Maverick
Last updated: August 10, 2026
Specs
| Vendor | Meta |
|---|
| Released | April 5, 2025 |
|---|
| Context window | 1M tokens |
|---|
| Input | $0.2 / M tokens |
|---|
| Output | $0.8 / M tokens |
|---|
Benchmarks
| mmlu | 85.1 |
|---|
| gpqa | 54.3 |
|---|
| humanEval | 82 |
|---|
| arenaElo | 1322 |
|---|
| sweBench | 45 |
|---|
| sweBenchPro | 35 |
|---|
| liveCodeBench | 58 |
|---|
| aiderPolyglot | 55 |
|---|
| terminalBench | 28 |
|---|
| mmluPro | 72 |
|---|
| hle | 8 |
|---|
| aime | 60 |
|---|
| math500 | 84 |
|---|
| mmmu | 66 |
|---|
Strengths
- Open weights, self-host on your own hardware
- Strong multilingual base
- Very cheap when self-hosted
Weaknesses
- Needs serious GPU infrastructure to self-host
- Reasoning behind closed frontier models
Best for
- Privacy-sensitive industries (legal, healthcare, defence)
- On-premise deployments
- Fine-tuning on proprietary data
Compare with other AI models