Meta: Llama 3.3 70B Instruct vs Microsoft: Phi 4
Meta: Llama 3.3 70B Instruct is cheaper than Microsoft: Phi 4 by about 100% on a blended (75% input / 25% output) basis. Meta: Llama 3.3 70B Instruct charges $0.000000 / $0.000000 per 1M input/output tokens; Microsoft: Phi 4 charges $0.070000 / $0.140000.
| Field | Meta: Llama 3.3 70B Instruct | Microsoft: Phi 4 |
|---|---|---|
| Input $/Mtok | $0.000000 | $0.070000 |
| Output $/Mtok | $0.000000 | $0.140000 |
| Cache read $/Mtok | – | – |
| Cache write $/Mtok | – | – |
| Input $/Mchar | 0.000000 | n/a |
| Output $/Mchar | 0.000000 | n/a |
| Context window | 128000 | 16384 |
| Tokenizer | llama3 | – |
Price history
Same workload, monthly cost
Illustrative monthly cost at 10M tokens/month, split per workload preset. Not a substitute for the calculator, which uses your own volume, cache and batch assumptions.
| Workload | Meta: Llama 3.3 70B Instruct | Microsoft: Phi 4 |
|---|---|---|
| Coding agent | $0.00 | $0.74 |
| RAG chat | $0.00 | $0.81 |
| Summarisation (batch) | $0.00 | $0.77 |
| Classification | $0.00 | $0.71 |
| Chat assistant | $0.00 | $0.98 |
| Content generation | $0.00 | $1.26 |
Cite this page
Meta: Llama 3.3 70B Instruct vs Microsoft: Phi 4: $0.000000/$0.000000 vs $0.070000/$0.140000 per 1M tokens, as of 2026-10-04 (Token Barometer, https://tokenbarometer.com/compare/meta_llama-3.3-70b-instruct-vs-microsoft_phi-4).
Meta: Llama 3.3 70B Instruct page | Microsoft: Phi 4 page | Open in calculator