Last refreshed 2026-05-19. Next refresh: weekly.
Why use Phi-2 on Replicate API?
Replicate API offers Phi-2 with pay-as-you-go pricing at $0.05/1M input tokens. Replicate is a cloud-based platform that enables users to run machine learning models easily and efficiently.
Compare Phi-2 across 5 providers to find the best fit for your use caseSetup recipe
Python + curlpip install replicateexport REPLICATE_API_TOKEN=...import replicate
output = replicate.run(
"lucataco/phi-2",
input={"prompt": "Hello"}lucataco/phi-2Request example
import replicate
# reads REPLICATE_API_TOKEN from env
# lucataco/phi-2 format: "owner/model-name" (latest version) or "owner/model-name:version-hash"
output = replicate.run(
"lucataco/phi-2",
input={"prompt": "Hello"}
)
# Output is a list or generator depending on the model
print("".join(output))Gotchas
- Use provider model ID "lucataco/phi-2", not the LLMReference slug "phi-2".
- Replicate uses "owner/model-name" format (e.g. "meta/meta-llama-3-8b-instruct") for the latest version, or "owner/model-name:version-sha" to pin to a specific version. The REST endpoint splits owner and model-name into the path: /v1/models/{owner}/{model-name}/predictions.
- The examples expect REPLICATE_API_TOKEN; rename it only if your application config maps the new variable.
Compare Phi-2 Across Providers
| Provider | Input (per 1M) | Output (per 1M) |
|---|---|---|
| Microsoft Foundry | $0.07 | $0.07 |
| Cloudflare Workers AI | — | — |
| Together AI | $0.10 | $0.10 |
| Fireworks AI | $0.10 | $0.10 |
| Replicate API | $0.05 | $0.25 |
Pricing
| Type | Price (per 1M) |
|---|---|
| Input tokens | $0.05 |
| Output tokens | $0.25 |
Capabilities
About Phi-2
Phi-2 is a compact language model by Microsoft endowed with 2.7 billion parameters and part of their Phi series. It shows formidable capabilities in reasoning and language understanding, outshining much larger models, even those with up to 25 times more parameters. Phi-2's training utilized a vast and diverse dataset of 1.4 trillion tokens, incorporating high-quality synthetic data and curated web content to bolster its common sense reasoning and general knowledge. Interestingly, despite lacking fine-tuning via reinforcement learning from human feedback (RLHF), it exhibits enhanced safety features and reduced bias.