Llama 4 Maverick
Language Available Full comparison ↗ meta/llama-4-maverick · by Meta
· mixture-of-experts
Pricing — 1 offering(s)
Input tokens
- $0.15 / 1M tokens (input) Historical 2025-04-05 → 2026-07-21
- $0.27 / 1M tokens (input) Current 2026-07-22 → present
Output tokens
- $0.60 / 1M tokens (output) Historical 2025-04-05 → 2026-07-21
- $0.85 / 1M tokens (output) Current 2026-07-22 → present
Showing the active price and any recorded history. Full pricing history is available via the paid API — see API docs.
See how Llama 4 Maverick fits into a cost-aware routing setup
See how →Capability profile
How Llama 4 Maverick rates across core capability dimensions, with the task-level evidence behind each rating.
Strong for its active-parameter cost; below frontier closed models on hardest tasks.
SWE-bench Verified (500 human-verified GitHub issues; patch must pass the repo's own tests; no file-location hints): resolves 21% — fails roughly 4 issues in 5. Independent (official SWE-bench leaderboard, maintainer-verified mini-SWE-agent submission; Meta's own channels report LiveCodeBench, not SWE-bench).
Function calling supported; less mature than Anthropic or OpenAI pipelines.
Good format compliance and instruction adherence after post-training.
1M-token context — the largest in its cost tier by a wide margin.
MoE efficiency: 17B active parameters deliver high throughput at low per-token cost.
Coding benchmarks
View the full coding benchmark leaderboard →| Benchmark | Score | Harness | Source |
|---|---|---|---|
| swe-bench-verified ↗ | 21.0% 2025-07 | mini-swe-agent 0.0.0 | 📊 leaderboard ↗ |
| livecodebench ↗ | 43.4% 2025-04 | — | 📊 leaderboard ↗ |
| aider-polyglot ↗ | 15.6% 2025-04 | aider | 📊 leaderboard ↗ |
| swe-bench-pro | 5.2% 2026-01 | swe-agent | 📊 leaderboard ↗ |
| terminal-bench-2-1 | 7.9% 2026-08 | terminus-2 | 🔬 independent ↗ |
| bigcodebench ↗ | 27.7% 2025-04 | — | 📊 leaderboard ↗ |
Operator guidance
Route here for the longest context available at this price point, or for vision tasks where GPT-4o cost is prohibitive. At the same $0.15/$0.60 price as GPT-4o mini, Maverick's open weights and 1M-token context are the differentiators. Step up to frontier closed models for the hardest reasoning tasks.
Use cases
- Ultra-long-context workloads (entire codebases, books, legal documents)
- Multimodal tasks (image + text) at budget cost
- Open-weights self-hosting where per-token cost must be minimised
Limitations
- Newer model; production stability still being established relative to GPT-4o mini
- Open weights — quality/price depends on the serving provider
- Capability ratings are qualitative, not a single cited benchmark run