What is Llama 3.3 70B?

Llama 3.3 70B is a proven AI model still widely deployed in production systems, developed by Meta and released in 2024-12. It handles general AI tasks including writing, analysis, summarisation, and question answering.

Is Llama 3.3 70B open-source or open-weight?

Llama 3.3 70B is an open-weight model. Its trained weights are publicly available for download, self-hosting, and fine-tuning. You can run it on your own infrastructure without per-token API fees. Note that open-weight is not the same as open-source: the training code and data may not be fully public.

What benchmarks has Llama 3.3 70B been evaluated on?

Llama 3.3 70B has been evaluated on: GPQA Diamond (50.5%), MMLU-Pro (68.9%), MATH-500 (77.0%), HumanEval pass@1 (88.4%). Scores tagged SR are self-reported by the provider; scores tagged 3P or AA have been independently verified by third parties.

Is Llama 3.3 70B available in FlowHunt?

Yes — Llama 3.3 70B is directly available in FlowHunt. You can use it to build AI agents, automated workflows, chatbots, and document processing pipelines without any additional setup.

How does Llama 3.3 70B compare to other AI models?

You can compare Llama 3.3 70B against all 24 models tracked on the FlowHunt AI Leaderboard using the sortable benchmark table. The Benchmark Rankings section on this page shows Llama 3.3 70B's position on each benchmark it has been evaluated on.

Llama 3.3 70B

EstablishedOpen-weightAvailable in FlowHunt

Try in FlowHunt

Llama 3.3 70B is a proven AI model still widely deployed in production systems, developed by Meta and released in December 2024. As an open-weight model, its trained weights are publicly available — you can self-host it, fine-tune it on proprietary data, or run it on-premise without API dependencies. It supports a 128K-token context window, adequate for most documents and multi-turn applications.

On the FlowHunt AI Leaderboard it is one of 24 tracked models, evaluated across 4 benchmarks. Standout scores: 88.4% on HumanEval (#1 of 2); 77.0% on MATH-500 (#3 of 4); 68.9% on MMLU-Pro (#7 of 8).

Llama 3.3 70B was released on December 6, 2024 as one of Meta’s final Llama 3 series models before the architectural transition to Llama 4. As a dense (non-MoE) 70B parameter model, it reflected the Llama 3 generation’s philosophy of straightforward scalable architecture: more parameters, better results, transparent structure. Its 128K context window and strong MMLU-Pro and SWE-bench scores made it the leading self-hosted model for teams with GPU resources adequate for dense 70B inference but not a MoE system. It remains popular in regulated industries — healthcare, finance, legal — where the transparency and auditability of a dense architecture, combined with open-weight availability, matter as much as raw capability.

Note: Near-405B quality at 70B cost. Text-only. IFEval 92.1%.

Try in FlowHunt

Released: December 2024
Parameters: 70B dense
Context: 128K tokens
Weights: Open
Tier: Established
Provider: Meta

Llama 3.3 70B Benchmark Rankings

Rank among all 24 models tracked on this leaderboard. SR = self-reported · 3P / AA / C = independently verified.

Benchmark	Score	Rank	Source	What it measures
GPQA Diamond	50.5%	13 of 14	SR	Graduate-level Google-proof science Q&A
MMLU-Pro	68.9%	7 of 8	SR	Hard knowledge reasoning, 12K questions
MATH-500	77.0%	3 of 4	SR	Hendrycks math (500 problems)
HumanEval pass@1	88.4%	1 of 2	SR	Python code generation (saturated at frontier)

Detailed Benchmark Scores

GPQA ◇ SR

50.5%

Graduate-level Google-proof science Q&A

MMLU-P SR

68.9%

Hard knowledge reasoning, 12K questions

MATH SR

77.0%

Hendrycks math (500 problems)

HumEval SR

88.4%

Python code generation (saturated at frontier)

Llama 3.3 70B vs. Established Peers

Head-to-head benchmark comparison with other Established-tier models. Higher is better for all metrics.

Benchmark	Llama 3.3 70B	GPT-4o	Llama 4 Maverick
GPQA ◇	50.5%	53.6%	69.8%
MMLU-P	68.9%	72.6%	80.5%
MATH	77.0%	—	—
HumEval	88.4%	—	—

About Meta

Meta Founded 2004 · Menlo Park, CA

Website →

Meta Platforms (formerly Facebook) is one of the world's largest technology companies, headquartered in Menlo Park, California, and best known for Facebook, Instagram, WhatsApp, and Threads. Through Meta AI (formerly FAIR), the company has become the most prolific publisher of open-weight large language models via the Llama family. The Llama 4 series — Maverick, Scout, and Behemoth — were released under permissive community licences and remain the most downloaded open-weight models on platforms like Hugging Face. Meta's AI strategy is distinct among major labs: rather than operating a commercial API, Meta releases models openly for research and commercial use, integrates generative AI features across its own applications, and has publicly advocated for open-weight releases as the most democratic approach to AI development.

Use Cases

What to Use Llama 3.3 70B For

Scientific & Technical Reasoning: Llama 3.3 70B achieves 50% on GPQA Diamond (#13 of 14 models). It handles general scientific Q&A at a solid level, though frontier models score 25+ percentage points higher on the hardest graduate-level problems.
Self-Hosted & Private Deployments: Llama 3.3 70B is open-weight (70B dense parameters) — its trained weights are publicly available for download and self-deployment. Run it on your own GPU hardware or private cloud to ensure data never leaves your environment, eliminate per-token API costs at scale, or fine-tune the model on proprietary datasets.

Strengths & Limitations

Strengths

Self-hosted & data-private deployments (open-weight)
Long-context tasks (128K context window)

Limitations

Below-frontier hard science reasoning (50% GPQA — frontier is 90%+)
Significantly behind 2026 frontier models on agentic coding and complex reasoning

More to Compare

Related AI Models

Browse the Full AI Model Leaderboard

Compare all 24 models across 11 benchmarks — sortable, sourced, and updated June 2026.

← Back to AI Leaderboard Try in FlowHunt

Frequently asked questions

: Llama 3.3 70B is a proven AI model still widely deployed in production systems, developed by Meta and released in 2024-12. It handles general AI tasks including writing, analysis, summarisation, and question answering.
: Llama 3.3 70B is an open-weight model. Its trained weights are publicly available for download, self-hosting, and fine-tuning. You can run it on your own infrastructure without per-token API fees. Note that open-weight is not the same as open-source: the training code and data may not be fully public.
: Llama 3.3 70B supports a 128K-token context window, adequate for most documents and multi-turn applications.
: Llama 3.3 70B has been evaluated on: GPQA Diamond (50.5%), MMLU-Pro (68.9%), MATH-500 (77.0%), HumanEval pass@1 (88.4%). Scores tagged SR are self-reported by the provider; scores tagged 3P or AA have been independently verified by third parties.
: Yes — Llama 3.3 70B is directly available in FlowHunt. You can use it to build AI agents, automated workflows, chatbots, and document processing pipelines without any additional setup.
: You can compare Llama 3.3 70B against all 24 models tracked on the FlowHunt AI Leaderboard using the sortable benchmark table. The Benchmark Rankings section on this page shows Llama 3.3 70B's position on each benchmark it has been evaluated on.

Learn more

Llama 4 Scout — Benchmarks, Capabilities, and Use Cases

Llama 4 Scout is a open-weight AI model from Meta (17B/109B MoE (16 experts)), classified as Established tier with a 10M tokens context window. It is available ...

Apr 5, 2025 4 min read

Llama 4 Maverick — Benchmarks, Capabilities, and Use Cases

Llama 4 Maverick is a open-weight AI model from Meta (17B/400B MoE (128 experts)), classified as Established tier with a 1M tokens context window. Explore bench...

Apr 5, 2025 4 min read

Mistral Large 3 — Benchmarks, Capabilities, and Use Cases

Mistral Large 3 is a open-weight AI model from Mistral (41B/675B MoE), classified as Advanced tier with a 256K tokens context window. It is available directly i...

Dec 2, 2025 4 min read