Home › Models › Llama 3.3 70B
Llama 3.3 70B API pricing
Llama 3.3 70B is a model from Meta (Llama) with a 128,000-token context window. Rates below are per 1,000,000 tokens, verified November 2026.
Price list
| Token class | Rate / 1M | Notes |
|---|---|---|
| Input | $0.88 | Prompt tokens not served from cache |
| Cached input | $0.088 | 90% below base input |
| Cache write |  | Not billed per token |
| Output | $0.88 | 1.0× the input rate |
| Batch API | −50% | Input and output, results within 24 h |
Open weights. Previous generation, still widely deployed.
What Llama 3.3 70B costs at three real workloads
Sticker prices rarely match the invoice. These three workloads use the rates above, with cache writes charged once per job and the batch discount applied only where the provider offers it.
| Workload | Requests | Tokens | Estimated cost |
|---|---|---|---|
| Support chat | 1,000 | 4K in / 600 out | $4.05 |
| RAG with caching | 1,000 | 50K in (45K cached) / 1K out | $9.28 |
| Nightly batch job | 10,000 | 10K in / 2K out | $52.80 |
Caching the 45K-token prefix saves about 79% on the RAG workload versus sending it uncached every call ($44.88 → $9.28). Batch mode halves the nightly job.
Where Llama 3.3 70B sits in the market
Out of 37 models in this dataset, Llama 3.3 70B ranks #16 cheapest by input rate, #8 by output rate, and #13 on a blended 1M-input-plus-250K-output workload.
| Model | Provider | Input | Output | Blended |
|---|---|---|---|---|
| Llama 3.3 70B | Meta (Llama) | $0.88 | $0.88 | $1.10 |
| Qwen 3 235B | Alibaba (Qwen) | $0.50 | $2 | $1 |
| Gemini 3.8 Flash | Google Gemini | $0.75 | $3.75 | $1.69 |
| Gemini 3.5 Flash-Lite | Google Gemini | $0.30 | $2.50 | $0.925 |
Blended cost = 1M input tokens + 250K output tokens at list price.
FAQ
How much does Llama 3.3 70B cost per 1M tokens?
$0.88 per 1M input tokens and $0.88 per 1M output tokens on the standard tier.
Is prompt caching cheaper on Llama 3.3 70B?
Yes. Cached input bills at $0.088 per 1M tokens, 90.0% below the $0.88 base input rate, with no separate cache-write rate published.
Does Llama 3.3 70B support the Batch API?
Yes. Batch requests bill at roughly half the standard input and output rates: the nightly job above costs $106 at list price and about $52.80 in batch mode.
Keep reading
- All Meta (Llama) models and rates
- Every model in the dataset, ranked by price
- Llama 4 Maverick vs Llama 3.3 70B
- GPT-6 Astra vs Llama 3.3 70B
- Claude Fable 5.1 vs Llama 3.3 70B
- Calculate your own monthly cost with the free estimator
Run these numbers in the calculator