Llama 3.3 70B Instruct
meta-llama/llama-3.3-70b-instructMeta’s open-weight, text-only 70B model for multilingual prompts, tool use, and long context.
131K tokens of context — about 98,304 words, on a scale that ends at 2M.
What it can do
- Tool use
- Open weights
- Takes text
- Gives text
Released 6 Dec 2024 · checked 3 Oct 2026.
About it
Llama 3.3 70B Instruct is Meta’s instruction-tuned, text-in/text-out language model. It has open weights, supports tools, and handles multilingual text. The model reads up to 131,072 tokens and can produce up to 16,384 tokens in one answer. Usage is priced separately for input and output tokens.
Review
In a line
Llama 3.3 70B Instruct is Meta’s 70-billion-parameter multilingual language model. It is instruction-tuned, so it is designed to follow written requests and return text.
The weights are open, and the model accepts and produces text only.
What it's good at
This model suits multilingual writing and other tasks that start with text and end with text. Its tool support can help software connect model responses to external actions when the surrounding software provides them.
Its open weights may suit people who want to work with the model in compatible software. See Meta models for related options.
Price in plain terms
Input costs $0.1 per million tokens, and output costs $0.32 per million tokens. Input tokens are what the model reads; output tokens are what it writes.
Reading a 500-word page costs under 0.01 cents. A chat turn with about 1,000 input tokens and 500 output tokens costs about 0.026 cents. 1,000 such turns cost about $0.26.
Compare it with cheapest language models if token cost is your main concern.
Context and limits
The context window is 131,072 tokens, meaning the model can consider a large amount of text in one request. It can write up to 16,384 tokens in one answer.
It handles text only.
Ways to use it
Use the open weights in compatible software, or choose a service that offers this model. Its model identifiers include meta-llama/Llama-3.3-70B-Instruct and meta-llama/llama-3.3-70b-instruct.
Tool support makes it relevant to software that needs model responses to trigger tools. See models with tool calling for alternatives.
Verdict
Choose Llama 3.3 70B Instruct when you need a text-only model with open weights, multilingual handling, tool support, and a very large context window. Its separate input and output rates also make usage easy to estimate for token-based workloads.
Look elsewhere if your work needs images or another modality, because this model is limited to text. It may also be more model than you need for short, simple prompts, so compare smaller or cheaper options before committing.
Verdict
Llama 3.3 70B Instruct fits projects that need long text handling, multilingual responses, and the flexibility of open weights. Tool support adds value when connected software can act on model output. At the stated example sizes, 1,000 chat turns cost about $0.26. It is not suitable for image tasks, since its inputs and outputs are text. Buyers should also check that their chosen software or service can run or provide this specific model before planning a workflow around it.
- Open weights are available.
- Supports multilingual text.
- Tool support is included.
- Reads up to 131,072 tokens.
- Input and output rates are listed separately.
- Text only; no image modality is listed.
- Maximum output is 16,384 tokens.
- Requires compatible software or a service to use the weights.
- Pick it ifLong, multilingual text workflows; Users wanting open model weights; Tool-connected text applications
- Skip it ifYou need image input or image output.; You want a fixed monthly subscription instead of token billing.; Your software cannot run or provide this model.
Pricing
Usage is billed by tokens: $0.1 per million input tokens and $0.32 per million output tokens. At the stated example sizes, 1,000 chat turns cost about $0.26.
Questions
Who makes Llama 3.3 70B Instruct?
Meta makes it. It belongs to the meta-llama family.
Is Llama 3.3 70B Instruct open source?
Its weights are open, so compatible software can work with them.
How much text fits in one request?
Its context window is 131,072 tokens.
How long can one answer be?
The maximum output is 16,384 tokens.
What does it cost?
Input is $0.1 per million tokens and output is $0.32 per million tokens. A 1,000-turn example costs about $0.26.
Does it support tools?
Yes. Tool support is listed for the model, so compatible software can use that capability.
On these boards
More from Meta
- Muse Spark 1.3$1.25 in
- Muse Spark 1.3 Contributor$0.10 in
- Muse Spark 1.2 Contributor$0.10 in
- Muse Glimmer 30B$0.35 in
- Muse Spark 1.2$1.25 in
- Muse Spark 1.1$1.25 in
- Llama Guard 4 12B$0.18 in
- Llama 4 Maverick$0.19 in
- Llama 4 Scout$0.10 in
- Llama 3.2 1B Instruct$0.027 in
- Llama 3.2 3B Instruct$0.050 in
- Llama 3.1 70B Instruct$0.40 in
- Llama 3.1 8B Instruct$0.050 in