Meta Llama Models
Access Meta Llama 3.3 70B, 3.1 405B, and Llama 4 Scout via nRouter's unified gateway with real-time streaming, automated failovers, and built-in guardrails.
Access Meta's instruction-tuned Llama models through nRouter's unified gateway with a single virtual API key. All Llama models run under the same budgets, guardrails, and audit logs as every other catalog model, settling at exact provider list pricing with zero per-token markup.
| Model ID | Architecture | Context | Max Out | Primary Workload |
|---|---|---|---|---|
meta/llama-3.3-70b-instruct | Dense 70B | 128k | 4,096 | High-throughput reasoning and coding |
meta/llama-3.1-405b-instruct | Dense 405B | 128k | 4,096 | Frontier logic and dataset distillation |
meta/llama-4-scout-17b-16e-instruct | MoE (17B active) | 128k | 8,192 | Multimodal vision and fast agent loops |
Calling Llama Models
Call Llama models via POST /v1/chat/completions using your virtual key: Llama 4 Scout accepts image inputs via standard image_url parts (Send Image Input). Set "stream": true for SSE streaming; attach guardrails with nrouter_guardrails (Per-Request Options). Requests settle at provider list price with 0% platform fee, reported in x-nr-request-cost.
Frequently asked questions
Which Meta Llama models can I call?
meta/llama-3.3-70b-instruct, meta/llama-3.1-405b-instruct, and multimodal meta/llama-4-scout-17b-16e-instruct.
How am I billed for Llama requests?
At exact provider list price with zero markup; unused credit holds are returned immediately upon completion.
Can Llama 4 Scout take image inputs?
Yes; Scout accepts images via image_url message parts alongside text.
Can I add guardrails to a single Llama request?
Yes; pass guardrail identifiers in nrouter_guardrails to evaluate them alongside standing organization rules.