Skip to content
Meta

Llama 3.3 70B Instruct

Llama 3.3 70B Instruct is Meta's refined text-only model. It targets 405B-class results at 70B serving cost, with improved instruction following and multilingual capability. Your use is subject to Meta's Terms & Privacy Policies.

View API reference
Input and output price
Input $0.72, Output $0.72, Per 1M tokens
24h uptime
Loading AI Gateway uptime
Get API keyRead the docs
import { streamText } from 'ai'


const result = streamText({
  model: 'meta/llama-3.3-70b',
  prompt: 'Why is the sky blue?'
})
Read docs

Copy link to headingPlayground

Try out Llama 3.3 70B Instruct by Meta. Usage is billed to your team at API rates. Free users (those who haven't made a payment) get $5 of credits every 30 days.

meta logo
meta logo

Llama 3.3 70B Instruct

Copy link to headingProviders

Route requests across multiple providers. Copy a provider slug to set your preference. Visit the docs for more info. Using a provider means you agree to their terms, listed under Legal.

Checking availability for your team
Provider
Context
Max Output
Latency
Throughput
Input
Output
Cache
Web Search
Capabilities
ZDR
No Training
Free Tier
Release Date
Bedrock
Legal:Terms•Privacy
128K8K0.2 s171 tps
$0.72/M
$0.72/M
—
12/06/2024

Copy link to headingUptime

Direct request success rate on AI Gateway and per-provider. Visit the docs for more info.

Copy link to headingThroughput

P50 throughput on live AI Gateway traffic, in tokens per second (TPS). Visit the docs for more info.

Copy link to headingLatency

P50 time to first token (TTFT) on live AI Gateway traffic, in milliseconds. View the docs for more info.

Getting started

Call Llama 3.3 70B Instruct through AI Gateway with the AI SDK generateText and streamText functions, or through the OpenAI Chat Completions, OpenAI Responses, and Anthropic Messages APIs by changing the base URL. AI Gateway authenticates the request and routes it to an available provider.

Install the AI SDK (pnpm add ai dotenv), create an API key from the API Keys page, and set it as AI_GATEWAY_API_KEY in your environment. Full setup is covered in the text generation quickstart.

index.ts
import { generateText } from 'ai';
import 'dotenv/config';


async function main() {
  const result = await generateText({
    model: 'meta/llama-3.3-70b',
    prompt: 'Why is the sky blue?',
  });


  console.log(result.text);
}


main().catch(console.error);

Top-level parameters

The same Llama 3.3 70B Instruct request in each API format AI Gateway supports.

top-level-params.ts
import { generateText } from 'ai';
import 'dotenv/config';


async function main() {
  const result = await generateText({
    model: 'meta/llama-3.3-70b',
    system: 'You are a concise technical assistant.',
    prompt: 'Summarize the tradeoffs between static generation and SSR.',
    maxOutputTokens: 1024,
    temperature: 0.5,
  });


  console.log(result.text);
}


main().catch(console.error);

Standard parameters like prompt, messages, temperature, and tools work as documented in the AI SDK docs. These are the parameters with model-specific behavior.

ParameterTypeRequiredDescription
modelstringYesModel ID in the form creator/model, e.g. meta/llama-3.3-70b. AI Gateway routes the request to an available provider.
maxOutputTokensnumberNoHard cap on generated tokens. Llama 3.3 70B Instruct supports up to 8,192 output tokens.
providerOptionsRecord<string, JSONValue>NoAI Gateway routing options under gateway, plus any provider-native options under the provider’s own namespace — see the table below.

Input limits

InputFormatsSourcesMax countMax sizeLimits
Text————Prompt and response share the 128K-token context window

Provider options

Set AI Gateway routing options under providerOptions.gateway. For provider-specific options, pass them under the provider’s namespace as documented by the AI SDK.

Learn more in the AI SDK provider docs.

provider-options.ts
import { generateText } from 'ai';
import 'dotenv/config';


async function main() {
  const result = await generateText({
    model: 'meta/llama-3.3-70b',
    prompt: 'Why is the sky blue?',
    providerOptions: {
      gateway: {
        only: ['bedrock'],
      },
    },
  });


  console.log(result.text);
}


main().catch(console.error);

These AI Gateway routing options apply to every model. Provider-specific options pass through under the provider’s own namespace (for example providerOptions.anthropic) exactly as documented by the AI SDK.

ParameterTypeRequiredDescription
providerOptions.gateway.onlystring[]NoRestrict routing to these provider slugs. Requests fail over only within the listed providers.
providerOptions.gateway.orderstring[]NoPreferred provider order. Listed providers are tried first; unlisted providers remain available as fallbacks.
providerOptions.gateway.sort'cost' | 'ttft' | 'tps'NoRank candidate providers by price, time to first token, or tokens per second instead of the default routing order.
providerOptions.gateway.zeroDataRetentionbooleanNoRoute only to providers with a zero-data-retention policy for this model.

Routing across providers

AI Gateway serves the same model through multiple providers and fails over automatically. order expresses a preference while keeping every provider eligible; only is a hard allowlist — if none of the listed providers are available the request fails instead of falling back.

Options under a provider's own namespace (for example providerOptions.anthropic) are forwarded to that provider with the request. Providers ignore option namespaces that don't apply to them, so it is safe to set provider options alongside gateway routing options.

Tool calling

Expose tools the model can call. Define each tool’s inputs with a Zod schema.

tool-calling.ts
import { generateText, tool } from 'ai';
import { z } from 'zod';
import 'dotenv/config';


async function main() {
  const result = await generateText({
    model: 'meta/llama-3.3-70b',
    prompt: 'What is the weather in San Francisco?',
    tools: {
      getWeather: tool({
        description: 'Get the current weather for a location',
        inputSchema: z.object({ location: z.string() }),
        execute: async ({ location }) => ({ location, temperatureC: 18 }),
      }),
    },
  });


  console.log(result.text);
}


main().catch(console.error);

Copy link to headingMore models by Meta

Model
Context
Latency
Throughput
Input
Output
Cache
Web Search
Capabilities
Providers
ZDR
No Training
Free Tier
Release Date
meta/muse-image-1.0
—
$0.01/img
—
meta logo
08/26/2026
meta/muse-glimmer-30b
131K0.2 s100 tps
$0.35/M
$1.50/M
Read$0.04/M
—
+1
fireworks logo
togetherai logo
08/10/2026
meta/muse-spark-1.2-contributor
1M3.8 s128 tps
$0.10/M
$0.20/M
Read$0.002/M
—
+2
meta logo
08/05/2026
meta/muse-spark-1.1
1M3.8 s203 tps
$1.25/M
$4.25/M
Read$0.15/M
—
+2
meta logo
07/09/2026
meta/muse-spark-1.3-contributor
1M3.0 s68 tps
$0.10/M
$0.20/M
Read$0.002/M
—
+2
meta logo
—
meta/muse-spark-1.3
1M3.0 s53 tps
$1.25/M
$4.25/M
Read$0.15/M
—
+2
meta logo
—

Copy link to headingAbout Llama 3.3 70B Instruct

Meta released Llama 3.3 70B Instruct on December 6, 2024 as the final model in its 2024 Llama release cadence. The 3.3 70B is text-only, but it represents a targeted refinement of the 70B tier. Llama 3.3 70B Instruct delivers similar performance to the 3.1 405B at a fraction of the serving cost.

The core improvements center on instruction following and multilingual capability. Instruction following (the model's ability to accurately execute detailed or constrained directions) is one of the most important capabilities in production deployments where system prompts encode complex behavioral rules. The multilingual improvements matter for enterprise applications serving global audiences: better handling of non-English instructions reduces the engineering overhead of maintaining separate language-specific prompts.

Llama Stack, which Meta standardized throughout 2024 as a set of interfaces for RAG and agentic applications, is fully compatible with the 3.3 70B. Teams already using Llama Stack distributions for toolchain orchestration can upgrade to the 3.3 generation without rearchitecting their integration layer.

Copy link to headingWhat To Consider When Choosing a Provider

  • Configuration: If you're migrating from Llama 3.1 70B, test your existing prompts against 3.3 70B before you switch. Improved instruction following can change output style enough that you'll adjust prompts. Compare $0.72 and $0.72.
  • Zero Data Retention: Zero Data Retention is available for this model. It is offered on a per-provider and model basis. See the documentation for details.
  • Authentication: AI Gateway authenticates requests using an API key or OIDC token. You do not need to manage provider credentials directly.

Copy link to headingWhen to Use Llama 3.3 70B Instruct

Best for

  • 405B quality at 70B cost: Applications that previously required the 405B for output quality but where serving a 400B+ parameter model is economically prohibitive
  • Complex system prompts: Workloads depending on precise instruction following for customer support bots, structured data extraction, and multi-step reasoning chains
  • Multilingual production deployments: Improved non-English instruction handling reduces prompt engineering overhead
  • Upgrading from 3.1 70B: Teams that want clear quality gains without moving to a larger, more expensive model

Consider alternatives when

  • Vision required: Image understanding is part of the task and 3.3 70B is text-only, so Llama 3.2 90B handles multimodal inputs
  • Maximum reasoning depth: Cost is not the constraint and Llama 3.1 405B remains the largest open model
  • Native multimodal architecture: Rather than adapter-based vision, consider Llama 4 Maverick or Scout

Copy link to headingConclusion

Llama 3.3 70B Instruct is the practical high-capability choice for organizations that need 405B-level instruction quality at 70B serving economics. Improved instruction following makes it well-suited to production systems with complex behavioral specifications.

Copy link to headingFrequently Asked Questions

  • What specifically improved in Llama 3.3 70B Instruct over Llama 3.1 70B?

    Instruction following quality and multilingual capabilities. Llama 3.3 70B Instruct delivers performance comparable to the much larger 3.1 405B, with refinements in how the model handles detailed and constrained instructions.

  • Is Llama 3.3 70B Instruct a drop-in upgrade from 3.1 70B?

    Architecturally, yes. But improved instruction following means outputs may differ in style or format compared to 3.1 70B for the same prompts. Run regression tests against existing prompts before switching production workloads.

  • Does Llama 3.3 70B Instruct support vision inputs?

    No. It is a text-only model. For multimodal workflows at the 70B scale, Llama 3.2 90B (adapter-based vision) or Llama 4 Maverick (natively multimodal) are the appropriate alternatives.

  • How does Llama 3.3 70B Instruct relate to the broader Llama ecosystem tooling?

    Fully compatible with Llama Stack distributions, which provide standardized interfaces for RAG and agentic application development.