Llama 4 Maverick vs DeepSeek V3.2: API Pricing Comparison
Meta Llama 4 Maverick vs DeepSeek DeepSeek V3.2: token prices, batch and cache rates, Arena ELO, and cost per request for 8 business use cases. Prices in USD per 1 million tokens. Data snapshot: 2026-09-06.
Llama 4 Maverick pricing (Meta)
- Input: $0.20 per 1M tokens
- Output: $0.70 per 1M tokens
- Batch API (input/output): not published
- Prompt-cache read: not published
- Context window: 1M
- Arena ELO: 1327 (as of 2026-09-01)
- Full Llama 4 Maverick pricing page
DeepSeek V3.2 pricing (DeepSeek)
- Input: $0.27 per 1M tokens
- Output: $0.40 per 1M tokens
- Batch API (input/output): not published
- Prompt-cache read: $0.13
- Context window: 164K
- Arena ELO: 1425 (as of 2026-09-01)
- Full DeepSeek V3.2 pricing page
Cost per request by business use case
List price vs optimized (prompt caching on the cached input share, batch API for async workloads, only where the provider publishes those rates).
| Use case | Tokens (in/out) | Llama 4 Maverick list | Llama 4 Maverick optimized | DeepSeek V3.2 list | DeepSeek V3.2 optimized |
|---|---|---|---|---|---|
| Support Ticket | 1,500 / 500 | $0.000648 | $0.000648 | $0.000604 | $0.000482 |
| Knowledge Q&A | 2,000 / 800 | $0.000957 | $0.000957 | $0.000858 | $0.000670 |
| Meeting Summary | 10,000 / 1,200 | $0.0028 | $0.0028 | $0.0032 | $0.0030 |
| Marketing Content | 2,500 / 1,800 | $0.0018 | $0.0018 | $0.0014 | $0.0013 |
| Coding Task | 3,000 / 2,000 | $0.0020 | $0.0020 | $0.0016 | $0.0014 |
| Invoice Processing | 1,500 / 600 | $0.000718 | $0.000718 | $0.000643 | $0.000583 |
| Call Summary | 2,000 / 700 | $0.000887 | $0.000887 | $0.000818 | $0.000791 |
| Agent Workflow | 6,000 / 3,000 | $0.0033 | $0.0033 | $0.0028 | $0.0022 |
Open this comparison in the interactive compare tool: add up to two more models, copy a shareable link, or export the table as CSV.
These are list prices: real cost also depends on caching, batching, negotiated discounts, marketplace agreements, and existing contracts. Get a FinOps review from OptimNow or compare all models.