Nvidia Nemotron 3.5 Lightning 30 B A3 B Bf 16 vs Llama 3.1.8 B Instruct Fp 8: cost comparison
Prices updated 2026-11-06 · computed from live provider data
Short answer: Llama 3.1.8 B Instruct Fp 8 is cheaper on a
3:1 input/output blend — $0.186 per million tokens versus
$0.362, a 49% difference.
Nvidia Nemotron 3.5 Lightning 30 B A3 B Bf 16 offers the larger context window at 65,536 tokens, against 32,000 for Llama 3.1.8 B Instruct Fp 8.
Head to head
| Nvidia Nemotron 3.5 Lightning 30 B A3 B Bf 16 | Llama 3.1.8 B Instruct Fp 8 | |
|---|---|---|
| Input, per 1M tokens | $0.150 | $0.152 |
| Output, per 1M tokens | $1.00 | $0.287 |
| Blended 3:1, per 1M | $0.362 | $0.186 |
| Context window | 65,536 | 32,000 |
| Vendor | gmi | cloudflare |
Monthly cost at real volumes
| Workload (input / output) | Nvidia Nemotron 3.5 Lightning 30 B A3 B Bf 16 | Llama 3.1.8 B Instruct Fp 8 |
|---|---|---|
| 1M in / 250K out | $0.400 | $0.224 |
| 10M in / 2.5M out | $4.00 | $2.24 |
| 100M in / 25M out | $40.00 | $22.38 |
Which should you pick?
Pick Nvidia Nemotron 3.5 Lightning 30 B A3 B Bf 16 when you need the longer context window (65,536 vs 32,000 tokens). Pick Llama 3.1.8 B Instruct Fp 8 when price per token dominates your workload — high-volume classification, extraction, or batch summarisation.