Migration guide · 5 minutes
DeepInfra → QuickSilver Pro
DeepInfra is already the budget option among DeepSeek resellers — QuickSilver Pro serves the latest V4 wave (V4 Flash for low-cost chat, V4 Pro for reasoning) with per-token cache pricing. Both APIs are OpenAI-compatible, so the move is a base-URL swap. For the full side-by-side analysis, see /vs/deepinfra.
- Code changes
- base_url · API key · model IDs
- Model IDs to rename
- 2 mappings
- Then
- test your core flows end-to-end
The steps
Sign up at quicksilverpro.io/dashboard. First top-up bonus: we match your first top-up 100%, up to $50 — added to your balance when you top up again (any amount from $5).
In your OpenAI SDK init, swap the base_url. Note DeepInfra's OpenAI-compatible path ends in /v1/openai.
- base_url="https://api.deepinfra.com/v1/openai" + base_url="https://api.quicksilverpro.io/v1"Replace your DeepInfra token with a QuickSilver Pro key.
- api_key=os.environ["DEEPINFRA_TOKEN"], + api_key=os.environ["QSP_KEY"],DeepInfra prefixes model IDs with the originating org. Drop the prefix and use the QuickSilver Pro short name.
DeepInfra QuickSilver Pro deepseek- ai/ DeepSeek- V4- Flash deepseek- v4- flash deepseek- ai/ DeepSeek- V4- Pro deepseek- v4- pro Run one representative request for each feature you use — chat, streaming, tool / function calling, and json_schema strict mode. Apart from the known difference below, any behavioral difference is a bug — report it.
One known difference: json_schema strict mode is model-dependent. The Claude models do not support it, so a schema there is advisory and the enforced path is a tool with
strict: true— structured output.
Full before/after
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.deepinfra.com/v1/openai",
api_key=os.environ["DEEPINFRA_TOKEN"],
)
r = client.chat.completions.create(
model="deepseek-ai/DeepSeek-V4-Flash",
messages=[{"role": "user", "content": "Hi"}],
)import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.quicksilverpro.io/v1",
api_key=os.environ["QSP_KEY"],
)
r = client.chat.completions.create(
model="deepseek-v4-flash",
messages=[{"role": "user", "content": "Hi"}],
)import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.deepinfra.com/v1/openai",
api_key=os.environ["DEEPINFRA_TOKEN"],
)
r = client.chat.completions.create(
model="deepseek-ai/DeepSeek-V4-Flash",
messages=[{"role": "user", "content": "Hi"}],
)What you'll pay after switching
Per 1M tokens, input / output. QuickSilver Pro rates vs DeepInfra's published per-token pricing.
| Model | QuickSilver Pro | DeepInfra | Savings |
|---|---|---|---|
| DeepSeek V4 Pro | $0.70 / $2.10 | $1.30 / $2.60 | ~19% output |
DeepInfra prices: no check date
Common migration pitfalls
Cached-input pricing changes the math
Both providers discount cached input. QuickSilver Pro bills cached-input tokens at a separate, lower cache-read rate on the DeepSeek V4 wave (and the Qwen/Kimi models). For workloads with a high cache-hit ratio, compare effective per-request cost — not list price alone — before switching.
Embeddings and audio don't move; image generation does
Embeddings and audio transcription stay on DeepInfra — keep both SDKs in your codebase if you use them. Image generation moves with you: Nano Banana Pro (Gemini 3 Pro Image), Nano Banana 2.1, FLUX.2 Pro, FLUX.1 Schnell, SDXL Turbo, FLUX.2 Klein, Qwen Image 3.0, Seedream 5.0 Pro, Seedream 4, Bria FIBO 1.5, Bria FIBO 2.0, GPT Image 2, GPT Image 2.5 Flare and GPT Image 2.5 Sunburst.
Model generation may differ
DeepInfra's undated V4 Flash and QuickSilver Pro's official 0731 revision are not the same build. V4 Pro is the same model family, but provider and quantization differences can still affect output style. Re-run evals after switching.
Rate limits work differently
QuickSilver Pro applies per-key throughput caps (default 600 req/min, 1M tok/min, 8–100 parallel per account by top-up tier). For bursty traffic, enable retry-on-429 and request a higher limit if needed.
Migrating from DeepInfra — FAQ
On the same V4 Pro model, QuickSilver Pro is $0.70 / $2.10 per 1M tokens versus DeepInfra's current $1.30/$2.60 rate — about 46% lower on input and 19% lower on output. DeepInfra's lower V4 Flash listing is an undated build, while QSP pins the official 0731 agent revision, so those rows are not a like-for-like price comparison.
Two lines: swap base_url to api.quicksilverpro.io/v1, use a new API key, and rename DeepInfra's current V4 model IDs to deepseek-v4-flash or deepseek-v4-pro.
Yes — cached-input tokens bill at a separate, lower cache-read rate on the DeepSeek V4 wave and the Qwen/Kimi models, so repeat prompts cost less than fresh input. Both providers discount cached input; benchmark effective per-request cost if cache-hit ratio is material for your workload.
Embeddings and audio transcription are not offered. Image generation is: Nano Banana Pro (Gemini 3 Pro Image), Nano Banana 2.1, FLUX.2 Pro, FLUX.1 Schnell, SDXL Turbo, FLUX.2 Klein, Qwen Image 3.0, Seedream 5.0 Pro, Seedream 4, Bria FIBO 1.5, Bria FIBO 2.0, GPT Image 2, GPT Image 2.5 Flare and GPT Image 2.5 Sunburst. And 47 of our 86 models accept image input.
Start saving in 5 minutes
First top-up matched 100%, up to $50 in bonus credits, added on your next top-up. Keep your code on the OpenAI SDK — only the base URL and key change.