Open weightsMeta/10 trustVerified

Llama 3.3 70B

Name: Llama 3.3 70B
Rating: 7

Meta's refined 70B workhorse: 128K context, 8 languages, massive deploy base.

Visit official page →

About

Llama 3.3 70B Instruct (meta-llama/Llama-3.3-70B-Instruct), released December 6, 2024, is Meta's refined 70-billion-parameter chat model — matching Llama 3.1-405B-class quality on several benchmarks at a fraction of the serving cost, per Meta. 128K-token context, 8 languages (EN/FR/DE/HI/IT/PT/ES/TH), tool-use chat template. Weights are downloadable (gated, manual approval) under the Llama 3.3 Community License — custom, non-OSI: royalty-free including commercial use, but with an acceptable-use policy, attribution requirements, and a clause requiring >700M-MAU companies to seek a separate license. Describe as "open weights", never "open source".

Specs

Context window

128K tokens

Pricing

Self-host: free (Llama 3.3 Community License). Hosted inference via dozens of providers — per-token pricing varies; check live provider pages.

Free quota

No official API; self-hosting on your own hardware is free.

API access

Self-host (Transformers, vLLM, Ollama, llama.cpp) after gated approval. No official Meta API.

Weights

Open weights

Released

December 6, 2024

Benchmarks

Provider-reported (Meta, Dec 2024): Llama 3.1-405B-competitive on MMLU, HumanEval, MATH at lower cost; strong multilingual — provider-reported. Widely re-benchmarked by the community in 2025; treat 2024 comparisons as historical.

Provider-reported unless a source says otherwise. Benchmarks are marketing-adjacent by nature — treat them as directional.

Use cases

chatcodingmultilingualRAG

Trust score — 7/10

/10 trustVerified

Scored 7/10 (verified tier, metrics-v1). Evidence: HF: meta-llama/Llama-3.3-70B-Instruct; HF downloads 297,980 (peer-normalized); released 676 days ago. Manual review: Meta's refined 70B (128K, 8 langs, 3.1-405B-class quality per Meta); huge deploy base; superseded by Llama 4. License: Llama 3.3 Community License — custom, non-OSI (700M MAU clause, acceptable-use policy, attribution); describe as 'open weights', never 'open source'.

How the trust score is built

Full methodology →

Adoption · stars/downloads, log-scaled vs peers: 8.1/10
Maintenance · recency, releases, issue load: 5.0/10
Openness · license + docs: 7.2/10
Manual review · sanity read + red-flag check: 8.0/10

· HF: meta-llama/Llama-3.3-70B-Instruct
· HF downloads 297,980 (peer-normalized)
· released 676 days ago
· manual review: Meta's refined 70B (128K, 8 langs, 3.1-405B-class quality per Meta); huge deploy base; superseded by Llama 4