Llama 4 Maverick
Meta · released Apr 5, 2025
Llama 4 Maverick 17B Instruct (128E) is a high-capacity multimodal language model from Meta, built on a mixture-of-experts (MoE) architecture with 128 experts and 17 billion active parameters per forward...
Specification
- Context window
- 1.05M
- Max output
- 16K
- Knowledge cutoff
- Aug 31, 2024
- Parameters
- 401.6B
- Licence
- other
- Serving providers
- 4
- Moderated
- No
- Uptime
- 100.0%
Intelligence
9.3
39th percentile
Coding
16.3
Coding Index
Agentic
—
Agentic Index
Output speed
78 t/s
Median across providers
Latency
323ms
Time to first token
Cost per task
$0.03
Estimated
Benchmarks
Where the score comes from
The Intelligence Index is a composite. These are the underlying evaluations this model was actually measured on.
Evaluation scores
Percentage correct · higher is better
- MMLU-Pro80.9%
- GPQA Diamond67.1%
- AA-LCR (long context)50.0%
- IFBench43.0%
- LiveCodeBench39.7%
- SciCode31.7%
- AIME 202519.3%
- τ²-bench (Telecom)17.8%
- Terminal-Bench Hard6.8%
- Humanity's Last Exam4.9%
An evaluation missing from this list was not run for this model — it is not a zero.
View as table
| Evaluation | Score |
|---|---|
| MMLU-Pro | 80.9% |
| GPQA Diamond | 67.1% |
| AA-LCR (long context) | 50.0% |
| IFBench | 43.0% |
| LiveCodeBench | 39.7% |
| SciCode | 31.7% |
| AIME 2025 | 19.3% |
| τ²-bench (Telecom) | 17.8% |
| Terminal-Bench Hard | 6.8% |
| Humanity's Last Exam | 4.9% |
Against its peers
Intelligence Index · this model highlighted, nearest peers in grey
- R1 Distill Llama 70BDeepSeek9.8
- 9.6
- 9.6
- Llama 4 MaverickMeta9.3
- Qwen3 Coder NextQwen9.2
- Mistral Medium 3.1Mistral AI9.2
- gpt-oss-20bOpenAI9.0
- Mistral Medium 3Mistral AI9.0
Peers are the models sitting closest on the Intelligence Index — the set you would realistically choose between.
View as table
| Model | Intelligence |
|---|---|
| R1 Distill Llama 70B | 9.8 |
| Qwen3 Next 80B A3B Instruct | 9.6 |
| Qwen3 Coder 30B A3B Instruct | 9.6 |
| Llama 4 Maverick | 9.3 |
| Qwen3 Coder Next | 9.2 |
| Mistral Medium 3.1 | 9.2 |
| gpt-oss-20b | 9.0 |
| Mistral Medium 3 | 9.0 |
Percentile among all indexed models
Pricing
What it costs to run
List prices per million tokens, plus what one representative task works out to.
List price
- Input / 1M tokens
- $0.188
- Output / 1M tokens
- $0.652
- Cached input / 1M
- $0.05
- Blended 3:1
- $0.304
One task, estimated
$0.03
- Input tokens
- 50,000
- Output tokens
- 25,000
- Profile
- Standard
Estimated from list pricing: 50K input tokens plus 80K output tokens for reasoning models (25K for non-reasoning).
Serving providers
Speed and latency figures are medians across these providers, so a widely-served model reports a blend rather than any single endpoint.
Arena
Head-to-head generation quality
Elo from pairwise judgements, broken out by the kind of thing the model was asked to build.
- Overall Elo
- 901
- Win rate
- 40.8%
- Strongest at
- UI components
- Tournaments
- 184
Elo by category
Dot position on a 825–925 scale · Elo has no meaningful zero
- 901
- 901
- 873
- 840
Agent categories (full-stack apps, mobile apps) are only scored for models tested in agent harnesses.
View as table
| Category | Elo | Win rate |
|---|---|---|
| UI components | 901 | 40.8% |
| 3D scenes | 901 | 40.2% |
| Websites | 873 | 34.4% |
| Game development | 840 | 33.7% |