Meta

Llama-3.3-70B-Instruct

Generally availableOpen weights128K context

Summary

Llama-3.3-70B-Instruct is an open-weight language model from Meta, released on 6 Dec 2024. It accepts text and generates text, with a 128K-token context window and up to 8K output tokens per response. As of 18 Nov 2026, the lowest listed API price is $0.10 per million input tokens and $0.32 per million output tokens (DeepInfra), across 8 providers we track.

Open Llama instruction model for multilingual chat, reasoning, and coding

Last verified 18 Nov 2026. Every figure on this page links to its source.

Specifications

DeveloperMeta
Release date6 Dec 2024 [models.dev]
StatusGenerally available [models.dev]
WeightsOpen weights [Hugging Face]
Licensellama3.3 [Hugging Face]
Parameters70.6B [Hugging Face]
Context window128K (128,000 tokens) [models.dev]
Max output8K tokens [models.dev]
Knowledge cutoffDec 2023 [models.dev]
InputText [models.dev]
OutputText [models.dev]
Reasoning modeNo [models.dev]
Tool callingYes [models.dev]
Structured outputYes [models.dev]
Hugging Facemeta-llama/Llama-3.3-70B-Instruct [Hugging Face]

Llama-3.3-70B-Instruct API pricing by provider

USD per million tokens. Sorted by input price.

ProviderInputOutputCached inputContextSource
DeepInfra
meta-llama/Llama-3.3-70B-Instruct-Turbo
$0.10$0.32–131Kmodels.dev, 18 Nov 2026
Novita AI
meta-llama/llama-3.3-70b-instruct
$0.14$0.40–12Kmodels.dev, 18 Nov 2026
Cloudflare Workers AI
@cf/meta/llama-3.3-70b-instruct-fp8-fast
$0.29$2.25–24Kmodels.dev, 18 Nov 2026
Hugging Face Inference Providers
meta-llama/Llama-3.3-70B-Instruct
$0.59$0.79–131Kmodels.dev, 18 Nov 2026
Azure AI Foundry
llama-3.3-70b-instruct
$0.71$0.71–128Kmodels.dev, 18 Nov 2026
OpenRouter
meta-llama/llama-3.3-70b-instruct
$0.71$0.71$0.71128KOpenRouter, 18 Nov 2026
Vercel AI Gateway
meta/llama-3.3-70b
$0.72$0.72–128Kmodels.dev, 18 Nov 2026
Snowflake Cortex
snowflake-llama3.3-70b
–––128Kmodels.dev, 18 Nov 2026

Running Llama-3.3-70B-Instruct locally

Estimated memory, from the parameter count on Hugging Face. Includes about 0.8 GB of runtime overhead.

QuantizationWeightsTotal, 8K contextTotal, 32K context
Q4_K_M (4-bit)43 GB44 GB–
Q8_0 (8-bit)75 GB76 GB–
FP16 / BF16 (16-bit)141 GB142 GB–

At 4-bit with an 8K conversation, Llama-3.3-70B-Instruct needs about 44 GB, more than any single consumer graphics card; it needs a unified-memory machine, a multi-GPU setup or a split between graphics card and system RAM. The model's architecture is not public, so the conversation cache is not included. Check your own hardware, or see Best computer for local AI on our sister site AIPCs.

Frequently asked questions

What is the context window of Llama-3.3-70B-Instruct?

Llama-3.3-70B-Instruct has a context window of 128,000 tokens (128K) and can generate up to 8,192 tokens in a single response, according to models.dev.

How much does the Llama-3.3-70B-Instruct API cost?

Llama-3.3-70B-Instruct is listed from $0.10 per million input tokens and $0.32 per million output tokens (DeepInfra). Prices last verified 18 Nov 2026.

Is Llama-3.3-70B-Instruct open source?

Llama-3.3-70B-Instruct is an open-weight model: its weights are published on Hugging Face as meta-llama/Llama-3.3-70B-Instruct under the llama3.3 license. Check the license terms before commercial use.

When was Llama-3.3-70B-Instruct released?

Meta released Llama-3.3-70B-Instruct on 6 Dec 2024, according to models.dev.

Which providers offer Llama-3.3-70B-Instruct?

We track Llama-3.3-70B-Instruct on 8 providers: DeepInfra, Novita AI, Cloudflare Workers AI, Hugging Face Inference Providers, Azure AI Foundry, OpenRouter, Vercel AI Gateway, Snowflake Cortex.

How much VRAM does Llama-3.3-70B-Instruct need?

About 44 GB at 4-bit (Q4_K_M) with an 8K-token conversation, and about 76 GB at 8-bit. That is more than any single consumer graphics card holds. These are estimates from the model's published parameter count.

Similar models from other labs

Comparable on output type, weights, release date, price and context window. Prices are the lowest listed, USD per million tokens.

ModelLabReleasedContextInput $/1MOutput $/1MWeights
Phi-4-miniMicrosoft11 Dec 2024128K$0.07$0.30Open
Qwen2.5-Coder-32B-InstructAlibaba (Qwen)12 Nov 2024131K$0.06$0.20Open
Mistral Small 3.1 24BMistral AI17 Mar 2025128K$0.10$0.30Open
Command R7BCohere13 Dec 2024128K$0.04$0.15Open
Gemma 3 27B ITGoogle DeepMind12 Mar 2025131K$0.08$0.16Open
Command RCohere30 Aug 2024128K$0.15$0.60Open

Change history

  • Max output changed from 4K to 8K
  • Context window changed from 131K to 128K
  • Context window changed from 12K to 131K
  • Max output changed from 12K to 4K
  • Max output changed from 24K to 12K
  • Context window changed from 24K to 12K
  • Name changed from Llama 3.3 70B Instruct fp8 Fast to Llama-3.3-70B-Instruct
  • Structured output changed from no to yes
  • Structured output changed from yes to no
  • Name changed from Llama-3.3-70B-Instruct to Llama 3.3 70B Instruct fp8 Fast
  • Max output changed from 12K to 24K
  • Context window changed from 12K to 24K
  • Structured output changed from no to yes
  • Name changed from Llama 3.3 70B Instruct fp8 Fast to Llama-3.3-70B-Instruct
  • Context window changed from 24K to 12K
  • Max output changed from 24K to 12K
  • Context window changed from 131K to 24K
  • Structured output changed from yes to no
  • Max output changed from 16K to 24K
  • Name changed from Llama 3.3 70B Turbo to Llama 3.3 70B Instruct fp8 Fast
  • Context window changed from 128K to 131K
  • Name changed from Llama-3.3-70B-Instruct to Llama 3.3 70B Turbo
  • Max output changed from 33K to 16K
  • Max output changed from 8K to 33K
  • OpenRouter price changed: input $0.59 → $0.71, output $0.79 → $0.71 per 1M tokens

Sources