Metaopen weightsLlama 3.3 70B Instruct pricing
$0.10 per 1M input tokens and $0.32 per 1M output tokens. In real work that is 0.05¢ per chat answer and 0.21¢ to summarize a 20-page PDF.
What Llama 3.3 70B Instruct costs per task
Token counts estimated from a measured model of the same lab (Qwen3.8 Flash), rescaled by tokenizer. Reasoning tokens are billed as output.
| Task | Cost | Tokens used | Price rank | Basis |
|---|---|---|---|---|
| Answer a chat question | 0.05¢$0.52 / 1,000 | 87 in · 683 out · 909 reasoning | 63 of 336 | estimated |
| Reply to a customer email | 0.03¢$0.25 / 1,000 | 323 in · 97 out · 587 reasoning | 69 of 336 | estimated |
| Write a 1,500-word blog post | ≥0.51¢$5.13 / 1,000 | 103 in · 0 out · 16,000 reasoning | 133 of 336 | estimated |
| Write 10 product descriptions | ≥0.52¢$5.15 / 1,000 | 342 in · 0 out · 16,000 reasoning | 155 of 336 | estimated |
| Translate 1,000 words | 0.06¢$0.55 / 1,000 | 1,375 in · 1,156 out · 136 reasoning | 43 of 336 | estimated |
| Proofread one page (500 words) | 0.07¢$0.66 / 1,000 | 770 in · 665 out · 1,159 reasoning | 75 of 336 | estimated |
| Summarize a 20-page PDF | 0.21¢$2.15 / 1,000 | 13,015 in · 497 out · 2,141 reasoning | 71 of 336 | estimated |
| Extract data from an invoice | 0.07¢$0.75 / 1,000 | 497 in · 556 out · 1,617 reasoning | 81 of 336 | estimated |
| Analyze a 200-row spreadsheet | ≥0.58¢$5.75 / 1,000 | 6,306 in · 0 out · 16,000 reasoning | 79 of 336 | estimated |
| Answer from 5 retrieved documents (RAG) | 0.13¢$1.29 / 1,000 | 3,721 in · 966 out · 1,912 reasoning | 77 of 336 | estimated |
| Classify 100 support tickets | 0.28¢$2.84 / 1,000 | 1,451 in · 529 out · 7,879 reasoning | 105 of 336 | estimated |
| Review 500 lines of code | ≥0.56¢$5.60 / 1,000 | 4,835 in · 0 out · 16,000 reasoning | 98 of 336 | estimated |
| Write a SQL query | 0.09¢$0.86 / 1,000 | 175 in · 401 out · 2,225 reasoning | 88 of 336 | estimated |
| Run a 50-step coding agent | 25¢$255 / 1,000 | 2,241,975 in · 15,030 out · 80,850 reasoning | 123 of 336 | modeled |
Price at every provider
List prices per 1M tokens (input / output) at each host serving Llama 3.3 70B Instruct through OpenRouter today. Hosts can differ in quantization, context and speed.
| Provider | Input / output | Cached input | Context | Uptime |
|---|---|---|---|---|
 DeepInfrafp8 | $0.10 / $0.32 | - | 131K | 97% |
 Novitabf16 | $0.14 / $0.40 | - | 12K | 97% |
 Parasailfp8 | $0.22 / $0.50 | $0.11 | 131K | 100% |
| A AkashMLfp8 | $0.20 / $0.52 | $0.10 | 131K | 100% |
| C CoreWeavefp16 | $0.71 / $0.71 | $0.71 | 128K | 100% |
 Googleunknown | $0.72 / $0.72 | - | 128K | |
Googleunknown | $0.72 / $0.72 | - | 128K | |
 Groqunknown | $0.59 / $0.79 | $0.29 | 131K | 100% |
 SambaNovaunknown | $0.45 / $0.90 | - | 131K | 100% |
 Togetherunknown | $1.04 / $1.04 | - | 131K | 100% |
| C Cloudflarefp8 | $0.29 / $2.25 | - | 24K | 100% |
Price history
First seen in our price history on Dec 11, 2024 at $0.13 input / $0.40 output. Since then the output price moved from $0.40 to $0.32 per 1M tokens (20 changes).
FAQ
How much does Llama 3.3 70B Instruct cost?
Llama 3.3 70B Instruct costs $0.10 per 1 million input tokens and $0.32 per 1 million output tokens (list price on Oct 18, 2026). In practice one chat answer costs about 0.05¢ and summarizing a 20-page PDF about 0.21¢.
Where is Llama 3.3 70B Instruct cheapest?
Among 11 providers we track, the lowest list price today is at DeepInfra: $0.10 input and $0.32 output per 1M tokens.
Is the Llama 3.3 70B Instruct cost per task measured or estimated?
Estimated: Llama 3.3 70B Instruct was not run directly. Tokens come from Qwen3.8 Flash and are rescaled by the tokenizer of Meta. Real usage can differ.