MODELS
Llama 3.3 70B
Open-weights 70B that matches Llama 3.1 405B quality at a fraction of the cost.
Specs
- Context window
- 128,000
- Max output
- 8,192
- Modalities
- text
- Tool use
- ✓
- Vision
- —
- Streaming
- ✓
- License
- llama-3.3-community
- Released
- 2024-12-06
Pricing
Meta's Llama 3.3 70B is a text-only instruction-tuned model that Meta claims matches the much larger Llama 3.1 405B on most benchmarks, with a 128K context and tool-calling support. It's released under the Llama Community License, so teams can self-host on a single 8×H100 node (or quantized on less). Built for chat, coding, and multilingual tasks across eight languages — no vision, no native audio.
Editor's verdict
Pick this when you need open weights and refuse to pay 405B-class hosting bills — it's the practical Llama default for self-hosters in late 2024 and 2025. It still trails GPT-4o and Claude 3.5 Sonnet on hard reasoning and coding, and lacks vision entirely, so it's not the right call if you need multimodal input or frontier benchmark scores. For RAG, agents, and on-prem chat, it's the strongest 70B-class option Meta has shipped.
Reviews
No reviews yet. Be the first.
Last updated: 2026-04-29