Simulating realistic Document summarization parameters (25,000 in / 600 out with 10% cache reuse). Llama 4 Maverick (400B MoE) delivers a 77% cost reduction over GPT-5.6 Terra.
Quick answer
For Document summarization, Llama 4 Maverick (400B MoE) is the lower-cost option at $0.012 per request versus $0.0527 for GPT-5.6 Terra, a modeled saving of 77%.
Method & trust
The comparison uses 25,000 input tokens, 600 output tokens, and 10% cache reuse for the selected workload. Pricing is applied per model, then scaled to monthly request volumes.
| Traffic Volume Tier | Llama 4 Maverick (400B MoE) Monthly | GPT-5.6 Terra Monthly | Monthly Savings by picking Llama 4 Maverick (400B MoE) |
|---|---|---|---|
| 1,000 reqs/mo (Dev/Testing) | $12.00 | $52.70 | Save $40.70 / mo |
| 10,000 reqs/mo (Small App) | $120.00 | $527.00 | Save $407.00 / mo |
| 100,000 reqs/mo (Growth Production) | $1,200.00 | $5,270.00 | Save $4,070.00 / mo |
| 1,000,000 reqs/mo (Scale SaaS) | $12,000.00 | $52,700.00 | Save $40,700.00 / mo |
Llama 4 Maverick (400B MoE) is 77% cheaper for Document summarization workloads. At standard Document summarization parameter ratios (25,000 input tokens, 600 output tokens, 10% cache hit), Llama 4 Maverick (400B MoE) costs $0.012 per request compared to $0.0527 on GPT-5.6 Terra.
Llama 4 Maverick (400B MoE) offers a context window of 1,000,000 tokens (max output: 16,384), while GPT-5.6 Terra offers 1,050,000 tokens (max output: 128,000).
At 100,000 requests per month, using Llama 4 Maverick (400B MoE) saves $4,070.00 every month (or $48,840.00 annually) compared to GPT-5.6 Terra.
Long-context models pay off here — compare price per 1M tokens at your true document size. Summarize once, store the result; don't re-summarize unchanged documents. For batch backfills, nightly jobs can use cache-friendly request ordering.