Llama·Meta

Llama 3.2 3B Instruct

meta/llama-3.2-3b

1 provider · cheapest €0.15/M via Amazon Bedrock

Try in Playground Copy code Docs

operationaltextstreaming

01·01

Overview

Modell-Beschreibung und Capability-Matrix.

About this model

Kleines, schnelles Open-Source-Modell von Meta (3B Parameter).

Capabilities matrix

streaming

tools—

json—

vision—

reasoning—

embedding—

02·02

Providers

1 EU-Provider mit Pricing, Throughput, Latency und Uptime — sortierbar. Pin-Button kopiert den X-CleverRouter-Provider-Header.

	Provider Model ID	Context	ZDR						Pin
Amazon Bedrock	eu.meta.llama3-2-3b-instruct-v1:0	eu-central-1	opt-in	€0.15	€0.15	—	—	99.99%

Pin kopiert X-CleverRouter-Provider-Header. Setze ihn an deinen Request, um Routing zu fixieren.

03·03

Code samples

Drop-in OpenAI-kompatibel — drehe an System-Prompt, Temperatur, Max-Tokens und Streaming, die Snippets aktualisieren sich live.

System prompt

User message

Temperature0.7

Max tokens1024

Stream response

use-llama-3.2-3b.ts

curl https://cleverouter.eu/v1/chat/completions \
  -H "Authorization: Bearer $CLEVERROUTER_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
  "model": "meta/llama-3.2-3b",
  "messages": [
    {
      "role": "system",
      "content": "You are a helpful assistant."
    },
    {
      "role": "user",
      "content": "Hallo aus der EU."
    }
  ],
  "temperature": 0.7,
  "max_tokens": 1024
}'

04·04

Performance

Throughput, TTFT, E2E-Latency und Tool-Error-Rate — pro Provider, p50 über 24h.

No performance data for this model yet. Once the first requests roll through, p50/p95 latency, throughput and error rate per provider will appear here.

Amazon Bedrock

eu-central-1

Throughput

—tok/s

p50 · 24h

TTFT

—ms

p50 · 24h

E2E Latency

—s

p50 · 24h

Tool Err Rate

—%

last 7d

Source · Aggregiert aus der requests Tabelle, letzte 24 h. p50/p95 via Postgres-Native PERCENTILE_CONT.

05·05

Uptime

30-Tage-Heatmap pro Provider, aggregierte Verfügbarkeit und Incident-Historie.

Aggregated uptime

99.99%

Last 30 days · all providers combined

≥ 99 %95–99 %< 95 %no data

Amazon Bedrock

eu-central-1

99.99%

last 30 days

Recent incidents · last 30 days

Keine Incidents in den letzten 30 Tagen.

06·06

Apps & use cases

What you can typically build with Llama models — ready-made snippets to get you started.

Agentic workflows

Tool-Calling, mehrstufige Plaene und Function-Routing — ideal für Agents, die Datenquellen anzapfen oder APIs orchestrieren.

const tools = [
  { type: 'function', function: { name: 'searchDocs', /* ... */ } },
];
const res = await client.chat.completions.create({
  model: 'meta/llama-3.2-3b',
  messages,
  tools,
  tool_choice: 'auto',
});

Customer support bot

Mit System-Prompt + RAG-Kontext eingespannt liefert das Modell konsistente Support-Antworten in der Markensprache des Kunden.

const completion = await client.chat.completions.create({
  model: 'meta/llama-3.2-3b',
  messages: [
    { role: 'system', content: BRAND_SYSTEM_PROMPT },
    ...retrievedContext,
    { role: 'user', content: userQuestion },
  ],
});

Code refactor assistant

Diff-aware Refactoring oder Test-Generierung — gut geeignet als Pair-Programmer-Layer in CI oder im IDE-Plugin.

const stream = await client.chat.completions.create({
  model: 'meta/llama-3.2-3b',
  stream: true,
  messages: [{ role: 'user', content: `Refactor this:\n\n${code}` }],
});

Structured extraction

JSON-Mode oder Tool-Calling für das Prüfen, Klassifizieren und Strukturieren von Freitext — Rechnungen, E-Mails, CRM-Felder.

const res = await client.chat.completions.create({
  model: 'meta/llama-3.2-3b',
  response_format: { type: 'json_object' },
  messages: [{ role: 'user', content: 'Extract: { ... }' }],
});