Skip to content

Cost reference

What it costs to fine-tune Llama 3.1 70B

The cheapest setup we compared is LoRA on 4 × L40S: about $0.59 per million training tokens, so $5.90 for 10 million and $59.01 for 100 million. Full fine-tuning costs about $10.44 per 10 million tokens on 8 × B200.

By Will Schulz · Checked 15 November 2026 · Prices as of November 2026

GPU-hours and cost by GPU and method

Per epoch over the token count shown; multiply by your number of epochs. GPU-hours are the same however many GPUs the job is split over.

GPU · methodGPUs for memoryTokens/s per GPUGPU-hours per 1M1.0M tokens10M tokens100M tokens
H100 SXM · LoRA$4.39/hr31,4020.20$0.87$8.70$86.99
H100 SXM · QLoRA$4.39/hr11,0930.25$1.12$11.15$112
H100 SXM · Full fine-tuning$4.39/hr16 (over 8)9350.30———
B200 · LoRA$7.99/hr13,1890.09$0.70$6.96$69.60
B200 · QLoRA$7.99/hr12,4870.11$0.89$8.92$89.23
B200 · Full fine-tuning$7.99/hr82,1260.13$1.04$10.44$104
L40S · LoRA$1.09/hr45130.54$0.59$5.90$59.01
L40S · QLoRA$1.09/hr24000.69$0.76$7.57$75.66
L40S · Full fine-tuning$1.09/hr27 (over 8)3420.81———
RTX 5090 · LoRA$1.39/hrDoesn't fit2970.94———
RTX 5090 · QLoRA$1.39/hrDoesn't fit2321.2———
RTX 5090 · Full fine-tuning$1.39/hrDoesn't fit1981.4———

Compared with fine-tuning APIs

OptionPer 1M tokens10M tokens100M tokens
Rented GPU: LoRA on L40S (our estimate)$0.59$5.90$59.01
Together AILlama 3.1 70B not listed——
OpenAI gpt-4.1-mini fine-tuning (closed model; closed to new users)$5.00$50.00$500

API prices are per training token processed (dataset tokens × epochs), seen on 15 November 2026. Together AI charges a minimum per job. OpenAI's pricing page says fine-tuning is closed to new users. Rented GPU estimates leave out setup, failed runs and idle time, which an API absorbs.

Your run

Pre-filled for Llama 3.1 70B: 10M tokens, 3 epochs, LoRA on an H100.

Estimated cost

$26.10

H100 SXM-hours

5.9

Time on 3 GPUs

2.0 hours

Tokens per second per GPU

1,402

30M tokens processed. Needs 3 × H100 SXM for memory at 2,048-token sequences. GPU-hours assume perfect scaling across GPUs.

Assumptions

  • Compute per token: 6 FLOPs per active parameter for full fine-tuning (forward plus backward) and about 4 for LoRA and QLoRA, which skip the frozen weights' gradients. Llama 3.1 70B uses 70.554B parameters per token.
  • Utilisation: 40% of each GPU's dense BF16 peak. torchtune measured 3,083 tokens a second for Llama 3.1 8B LoRA on one RTX 4090, about 60% on the same accounting, so well-tuned code can beat these figures.
  • QLoRA runs at 78% of LoRA's speed, the ratio torchtune measured (2,413 vs 3,083 tokens a second), because the 4-bit weights are dequantised on the fly.
  • Memory decides the GPU count, from our Llama 3.1 70B VRAM estimates at 2,048-token sequences. Consumer cards are counted as single GPUs only.
  • Prices: on-demand Runpod Secure Cloud rates (page dated 12 November 2026). Community Cloud and spot capacity cost less.

Frequently asked questions

How much does it cost to fine-tune Llama 3.1 70B?
About $0.59 per million training tokens at the cheapest setting we compared (LoRA on 4 × L40S, $1.09/hour per GPU), so about $5.90 for 10M tokens and $59.01 for 100M. Budget for several runs: the first rarely wins.
How long does it take to fine-tune Llama 3.1 70B on an H100?
LoRA processes about 1,402 tokens a second on one H100 at 40% utilisation, so 10M tokens take about 2.0 hours of GPU time across 3 H100s. Full fine-tuning needs 16 H100s for memory and about 1.5× the compute.
Is it cheaper to fine-tune Llama 3.1 70B through an API?
Together AI's November 2026 price list doesn't include Llama 3.1 70B, so the choice is rented GPUs or a hosted model. Rented GPU time costs about $5.90 per 10M tokens by our estimate, plus setup and idle time.
Should I use LoRA or full fine-tuning for Llama 3.1 70B?
Start with LoRA or QLoRA. Full fine-tuning Llama 3.1 70B needs about 1134 GB, so 16 × 80 GB GPUs with sharded optimiser state, and about 1.5× the compute per token. LoRA usually gets close for format, style and task adaptation; move to full fine-tuning only when evaluation shows LoRA has plateaued.

Sources

  1. [1]Hugging Face. meta-llama/Llama-3.1-70B-Instruct config.json
  2. [2]Meta. meta-llama/Llama-3.1-70B-Instruct model card
  3. [3]Meta. The Llama 3 Herd of Models (arXiv:2407.21783)
  4. [4]Hu et al., 2021. LoRA: Low-Rank Adaptation of Large Language Models (arXiv:2106.09685)
  5. [5]Dettmers et al., 2023. QLoRA: Efficient Finetuning of Quantized LLMs (arXiv:2305.14314)
  6. [6]Kaplan et al., 2020. Scaling Laws for Neural Language Models (arXiv:2001.08361)
  7. [7]PyTorch. torchtune: memory and throughput benchmarks (Llama 3.1 8B on RTX 4090)
  8. [8]Together AI. Fine-tuning pricing (seen 15 November 2026)
  9. [9]OpenAI. API pricing: fine-tuning (seen 15 November 2026)
  10. [10]Runpod. GPU pricing (updated 12 November 2026)

Try the cheaper levers first.

Prompt tuning and model search in Kojen cost API calls, not GPU-hours, and often close the gap before any fine-tuning.

Start training free