Llama 3.3 70B.
Meta's best-in-class open-weight LLM — 70B class.
1× Nvidia L40S · $0.28/hr.
Most-aggressive quantisation we have a working recommendation for. Lower precision = less VRAM = cheaper hardware, at a small accuracy cost.
Cheapest hosted endpoints.
| Provider | Access | $/M in | $/M out | |
|---|---|---|---|---|
| OpenRouter | api aggregator | $0.1 | $0.32 | Launch ↗ |
| Together AI | hosted inference | $0.88 | $0.88 |
What it's best at.
Speed across providers.
Tokens/sec and time-to-first-token measured against the same prompt template on each provider's API.
| Provider | Tokens/sec | TTFT | Total |
|---|---|---|---|
| OpenRouter | 23.5 | 1167 ms | 2212 ms |
Variants in the Llama family.
Llama 3.1 70B
Llama 3.1 70B — production workhorse, superseded by 3.3 but still widely depl...
Llama 3.1 8B
Meta's most popular open-weight small LLM — fits anywhere.
Llama 3.1 405B
Meta's largest open-weight LLM — dense 405B, frontier-class at launch.
Llama 3.2 90B Vision
Meta's largest vision-capable Llama.
Llama 3.2 11B Vision
Meta's open-weight multimodal LLM — vision + text in 11B.
Llama 3.2 3B
3B Llama — laptop-class chat + RAG.
Llama 3.2 1B
Meta's smallest Llama — mobile + on-device target.
Workloads.
Frequently asked.
How do I run Llama 3.3 70B?
Where can I access Llama 3.3 70B?
How much does it cost to run Llama 3.3 70B?
Is Llama 3.3 70B open-source or proprietary?
Cheapest hardware per quantisation.
Each row is one quantisation tier (the same weights compressed differently). Lower precision → lower VRAM → cheaper hardware, at the cost of small accuracy loss. $/hr refreshed hourly from each provider's API.
| Quantisation | Cheapest GPU config | Total VRAM | Live $/hr | tokens/sec | |
|---|---|---|---|---|---|
|
FP16
FP16 — half precision (default)
|
192 GB | — | — | Compare → | |
|
FP8
FP8 — 8-bit float (Hopper / Blackwell)
|
94 GB | — | — | Compare → | |
|
INT8
INT8 — 8-bit integer
|
96 GB | $0.61/hr | — | Compare → | |
|
INT4
INT4 — 4-bit integer (~4× VRAM saving)
|
48 GB | $0.28/hr | — | Compare → |
What it costs per month across providers.
Estimate your monthly bill for Llama 3.3 70B across every host that publishes per-token pricing. Slide your token volumes; the chart + table re-rank cheapest-first.
Cheapest provider on the left.
Total monthly cost — input + output tokens combined.
Bill breakdown.
| Provider | Monthly total | |
|---|---|---|
| $1.64 | Sign up ↗ | |
| $10.56 |
Rent the GPU instead of paying per token.
For an open-weights model like Llama 3.3 70B, you can rent a GPU and serve inference yourself. The math: cheapest GPU rental × 730 hours/month + your electricity rate × power draw.
Assumes the GPU runs 24/7 at ~85% utilisation. If your traffic is bursty, you'll pay less for the API and probably more for the GPU (idle hours still cost rental). The breakeven analysis lives on the Self-host vs API breakeven tool.
What it's best at.
Scores normalised against benchmark ceilings (100 = perfect). Coloured by tier — coral 80+ frontier, lavender 65+ strong, sage 50+ solid, slate below.
Published scores.
| Benchmark | Score | Source |
|---|---|---|
| GPQA | 50.5 | official ↗ |
| MATH | 77.0 | official ↗ |
| MMLU | 86.0 | official ↗ |
| IFEval | 92.1 | official ↗ |
| MMLU-Pro | 68.9 | official ↗ |
| HumanEval | 88.4 | official ↗ |
Independent rankings.
About Llama 3.3 70B.
Llama 3.3 70B is Meta's flagship open-weight model in the 3.3 series — competitive with Llama 3.1 405B on most benchmarks at 1/6th the parameter count, made possible by improved post-training. License permits commercial use under 700M MAU. Drop-in replacement for Llama 3.1 70B with the same tokenizer + same context window. Strong tool-use support; widely deployed on Hugging Face Inference, Together AI, Fireworks, Groq, and self-hosted via vLLM/TGI. Runs on 2× H100 (FP16) or 1× H100 (INT8/INT4).
How it's built.
How much it can remember.
What it can do.
Every place this model is hosted.
Self-hosted on rented GPU
self hostedTogether AI
hosted inferenceGroq
hosted inferenceIndustry-leading tokens/sec