# Llama 3.3 70B

> Most popular open model. Great all-rounder.

Human version: https://kymaapi.com/models/llama-3.3-70b
Live JSON: `GET https://kymaapi.com/v1/models` (no auth required)

## Facts

- **Model ID**: `llama-3.3-70b` — pass this as `model` in the request body
- **Creator**: Meta
- **Released**: 2024-12-06
- **Context window**: 128K tokens
- **Max output**: 8K tokens per response — a hard ceiling, not a default
- **Price**: $0.135 in / $0.432 out per 1M, cached input 10%
- **Capabilities**: tools, reasoning, caching
- **Alias**: `balanced` resolves to this model

## Call it

```bash
curl https://kymaapi.com/v1/chat/completions \
  -H "Authorization: Bearer $KYMA_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "llama-3.3-70b", "messages": [{"role": "user", "content": "Hello"}]}'
```

## Positioning

Llama 3.3 70B is a strong, cost-effective text-only model suited for general-purpose chat, code generation, and tool use. Reach for it when you need a reliable open-weight baseline that balances capability and throughput without premium pricing.

## About

Meta's 70B parameter model handles text-only workloads with a 128k context window and an 8192 token output limit. It supports structured outputs, function calling, and reasoning workflows, making it a practical default for text generation tasks.

On Kyma, this model resolves the `balanced` alias. Requests are routed through an OpenAI-compatible gateway with automatic failover, and the platform tracks exact costs in the `usage.cost` field while reporting the active model in the `X-Kyma-Model` header.

The model operates at a medium speed tier with a median throughput of 234 tokens per second. It does not support vision or multimodal inputs, and outputs are strictly capped at 8192 tokens.

## Use cases

- **General Chat Applications** — Handles conversational workloads with reliable context retention across 128k tokens.
- **Code Generation And Review** — Writes and debugs code while supporting structured outputs for IDE integrations.
- **Tool-Driven Agent Workflows** — Executes function calls and reasoning steps reliably for autonomous backend tasks.
- **Prompt-Cached System Prompts** — Reuses long system instructions efficiently since Kyma bills cached prefixes at 10% of the standard input rate.

## Not ideal for

Avoid this model if your workload requires image understanding, outputs longer than 8192 tokens, or demands the lowest possible latency.

## Pick something else when

- You need image or video input → `gemma-4-31b`
- You need faster throughput for high-volume requests → `deepseek-v4-flash`
- You need higher reasoning capability for complex tasks → `qwen-3.7-max`

## See also

- All models: https://kymaapi.com/models.md
- Pricing: https://kymaapi.com/pricing.md
- Other models by Meta: https://kymaapi.com/models?q=Meta