Llama 3.3 70B API pricing

15 providers serve Llama 3.3 70B. Llama 3.3 70B is an open-weight model (Llama 3.3 Community) with 70B total parameters, up to 131K context. Prices below are per 1M tokens, cheapest input first (self-hosting math further down).

Refreshed 2026-08-23 · list prices from provider APIs
Input floor $0.100/1M at OpenRouter Output floor $0.200/1M at Crusoe and Nscale Find your workload winner →
Provider Input $/1M Output $/1M Cache read $/1M Context
OpenRouter marketplace quote $0.100 $0.320 - 131K Use
Hyperbolic $0.120 $0.300 - 131K Use
Nebius $0.130 $0.400 - 128K Use
Novita AI $0.135 $0.400 - 131K Use
Crusoe $0.200 $0.200output floor - 131K Use
Nscale $0.200 $0.200 - - Use
DeepInfra $0.230 $0.400 - 131K Use
SambaNova $0.600 $1.20 - 131K Use
OVHcloud $0.670 $0.670 - 131K Use
Azure AI Foundry $0.710 $0.710 - 128K Use
IBM watsonx $0.710 $0.710 - 128K Use
Oracle OCI $0.720 $0.720 - 128K Use
Vercel AI Gateway $0.720 $0.720 - 128K Use
Cerebras $0.850 $1.20 - 128K Use
Scaleway $0.900 $0.900 - 128K Use

The cheapest way to run Llama 3.3 70B

On an illustrative 70M-input / 30M-output monthly workload, today's lowest listed cost is roughly $16.60/month, at OpenRouter. Input-heavy, output-heavy and cache-heavy workloads can produce different winners. Use the live mix calculator below rather than combining floors from two different providers. On output-token cost alone, the API floor remains below our self-hosting estimate of ~$0.376/1M. Details below. Run your own numbers in the breakeven calculator, or read the breakeven math.

Your actual monthly cost

Uses today's list prices

Cache read uses the listed cache rate where available; otherwise it falls back to normal input price. Batch, write-cache, volume and negotiated discounts are excluded.

Or run it yourself: Llama 3.3 70B self-hosting economics

Llama 3.3 70B is open-weight (Llama 3.3 Community), so the API price above competes with the GPU-hour market. 70B total parameters needs roughly 97 GB VRAM at FP8 or 49 GB at INT4, KV-cache headroom included.

GPU that fits (FP8, single card)VRAMFrom $/hrCheapest at
H200 NVL 141 GB $0.500community RunPod Rent
MI300X 192 GB $0.500community RunPod Rent
H200 141 GB $2.00spot Verda
B200 180 GB $3.06spot Verda
B300 288 GB $3.75spot Verda
GB300 288 GB $4.31spot Verda
Breakeven estimate: a well-batched vLLM deployment on one H200 NVL ($0.500/hr) at ~739 aggregate tok/s and 50% utilization works out to roughly $0.376 per 1M output tokens, versus $0.200 via the cheapest output-token API (Crusoe and Nscale). The API is cheaper unless your utilization or throughput beats these assumptions. Planning estimate: throughput varies with hardware, quantization, batch size and context. Tune it in the calculator.

Related models

ModelCheapest in $/1MCheapest out $/1MProviders
Llama 3.1 8B $0.020 $0.030 15
Llama 4 Scout $0.050 $0.100 11
Llama 3.1 70B Instruct $0.120 $0.300 10
Llama 4 Maverick $0.050 $0.100 9
Llama 3.2 3B $0.020 $0.020 9
Llama-3-70b $0.120 $0.300 7

FAQ

What is the cheapest API for Llama 3.3 70B?

As of 2026-08-23, the lowest input price for Llama 3.3 70B is OpenRouter at $0.100 per 1M input tokens. The lowest output price, $0.200 per 1M tokens, is likewise shared by Crusoe and Nscale. The cheapest provider for a real workload depends on its input/output mix. The next-lowest input price is Hyperbolic at $0.120, a 20% difference.

How much VRAM do you need to self-host Llama 3.3 70B?

Llama 3.3 70B has 70B parameters, so plan for roughly 97 GB of VRAM at FP8 or 49 GB at INT4/AWQ, KV-cache headroom included. A single H200 NVL (141 GB) fits it.

← All models · GPU rental prices · Breakeven calculator · Get this data as JSON