HopBaseThe short answer: HopBase serves DeepSeek V4.1 Flash 1017, the 2026-10-17 release, from a Singapore data center under its own model ID, deepseek-v4.1-flash-1017-sg. Call it like any OpenAI-compatible model: set the base URL to https://api.hop-base.com/v1, use a HopBase key that includes the "DeepSeek Singapore Data Center" group, and pass the ID with its -sg suffix. It has a 1M-token context window, up to 393,216 output tokens per call, and supports streaming, tool calling, thinking (reasoning_content) and automatic prefix caching. Its official list price and peak/off-peak schedule are the same as deepseek-v4.1-flash-1017. Singapore and Beijing are both on UTC+8, so the peak windows read the same on a Singapore clock: weekdays 09:00–12:00 and 14:00–18:00. Every other hour, weekends included, bills at half the peak rate. Billing is pay-as-you-go in USD with no subscription, and the same key can also include Claude, GPT or video groups.
Who this is for
Southeast Asian teams that already use DeepSeek, or want to, and prefer a deployment labelled by region: customer-support bots, document pipelines, coding agents, and SaaS products that want a low-cost reasoning model next to Claude or GPT on the same key.
The essentials
| Item | Value |
|---|---|
| Model ID | deepseek-v4.1-flash-1017-sg (the suffix is part of the ID; do not add or remove it) |
| Key group | DeepSeek Singapore Data Center, open to all users (see the DeepSeek docs) |
| Release | DeepSeek V4.1 Flash, 2026-10-17 release |
| Context / max output | 1M tokens / 393,216 tokens per call |
| Capabilities | Streaming, tool calling, thinking (reasoning_content), automatic prefix caching |
| Endpoints | POST /v1/chat/completions, POST /v1/responses and Anthropic Messages POST /v1/messages |
| Official list price (peak) | ¥2.00 input / ¥0.04 cached input / ¥8.00 output per 1M tokens, charged from your USD balance; see hop-base.com/pricing |
| Off-peak | All hours outside weekdays 09:00–12:00 and 14:00–18:00 Beijing time (= Singapore time), at half the peak rate |
A key that only has the main DeepSeek group gets a 404 for this ID. Add the DeepSeek Singapore Data Center group under API Keys in the console.
Off-peak hours on a Singapore clock
DeepSeek publishes its peak windows in Beijing time. Singapore Standard Time is also UTC+8, so there is nothing to convert:
| Singapore time (Mon–Fri) | Rate |
|---|---|
| 00:00–09:00 | Off-peak (half price) |
| 09:00–12:00 | Peak |
| 12:00–14:00 | Off-peak (half price) |
| 14:00–18:00 | Peak |
| 18:00–24:00 | Off-peak (half price) |
| Saturday and Sunday, all day | Off-peak (half price) |
A request is priced by the moment it finishes and always carries a single unit price: one that starts at 17:59 and ends at 18:01 is billed entirely at the off-peak rate. In practice, interactive traffic during office hours pays the peak rate, while batch work you can schedule (nightly summarisation, back-filling a search index, regenerating product descriptions) can run in the lunch window, after 18:00 or at weekends at half the rate.
Calling it from Python
import os
from openai import OpenAI
client = OpenAI(
base_url="https://api.hop-base.com/v1",
api_key=os.environ["HOPBASE_API_KEY"], # keep the key in an env var or secret manager
)
# Confirm the key can see the model before you send real traffic
ids = [m.id for m in client.models.list().data]
assert "deepseek-v4.1-flash-1017-sg" in ids, "Add the DeepSeek Singapore Data Center group to this key"
resp = client.chat.completions.create(
model="deepseek-v4.1-flash-1017-sg",
messages=[
{"role": "system", "content": "You answer in the customer's language: English, Malay, Bahasa Indonesia, Thai or Vietnamese."},
{"role": "user", "content": "Boleh saya tukar alamat penghantaran selepas membuat pesanan?"},
],
stream=True,
stream_options={"include_usage": True}, # needed to receive usage when streaming
)
for chunk in resp:
if chunk.choices and chunk.choices[0].delta.content:
print(chunk.choices[0].delta.content, end="")
if chunk.usage:
print("\n", chunk.usage) # prompt / completion tokens for your own reconciliation
Tools that speak the Anthropic Messages protocol (for example Claude Code or the Anthropic SDK) can use the same ID with the base URL https://api.hop-base.com, without /v1. Responses API clients use https://api.hop-base.com/v1.
Run batch jobs only in off-peak windows
from datetime import datetime
from zoneinfo import ZoneInfo
PEAK = [(9, 12), (14, 18)] # weekday peak hours, Beijing time = Singapore time (UTC+8)
def is_off_peak(now: datetime | None = None) -> bool:
now = now or datetime.now(ZoneInfo("Asia/Singapore"))
if now.weekday() >= 5: # Saturday, Sunday
return True
return not any(start <= now.hour < end for start, end in PEAK)
if is_off_peak():
run_nightly_batch() # your job
else:
queue_for_later()
The published schedule is defined only by weekday and hour, which is exactly what this helper checks.
One key for DeepSeek, Claude and GPT
A HopBase key can include 1 to 10 groups across platforms, so the Singapore DeepSeek group can sit on the same key as a Claude group, a GPT group or a video group. Each request is served and billed by the one group that covers the requested model and is never silently moved to another group. Call GET /v1/models with the key and use only the IDs it returns.
That makes a common Southeast Asian startup pattern easy: a low-cost reasoning model (deepseek-v4.1-flash-1017-sg) for high-volume support and extraction, and claude-sonnet-5-5 ($2 input / $10 output per million tokens at Anthropic's list price) or gpt-6-sol ($2 / $10 at OpenAI's list price) for the harder agentic steps, all behind one HOPBASE_API_KEY.
Which DeepSeek ID should you pick?
HopBase lists DeepSeek under several IDs, each in its own group:
deepseek-v4.1-flash-1017-sg: the DeepSeek Singapore Data Center group, described above.deepseek-v4.1-flash-1017anddeepseek-v4.1-flash: the main DeepSeek group, which also serves V4 Pro and V4 Flash.deepseek-v4.1-flash-us,deepseek-v4-pro-us,deepseek-v4-flash-us: the DeepSeek US Region group, billed at the vendor's US-region list prices.
Do not guess at latency from a region label. Measure time-to-first-token from your own servers before you decide; our reproducible TTFT method works against any provider.
If your project has data-residency or personal-data requirements (for example under Singapore's PDPA), contact an advisor before sending personal data through any model.
Paying and controlling spend
- Pay-as-you-go in USD with no subscription: Visa / Mastercard, Apple Pay, Google Pay, Alipay, AlipayHK and WeChat Pay. Any payment fee is shown on the top-up page before you pay.
- Give each customer or project its own key with its own spend cap and concurrency cap. A key that hits its spend cap returns
402 insufficient_quota; a concurrency cap returns429. - Requests rejected or failed before producing output are not billed. The console Usage page exports itemised CSV for line-by-line reconciliation.
FAQ
What is the model ID for DeepSeek in Singapore?
deepseek-v4.1-flash-1017-sg. The -sg suffix is part of the ID. Your key needs the "DeepSeek Singapore Data Center" group, and GET /v1/models should return the ID before you use it.
Is it the same model as deepseek-v4.1-flash-1017?
Both are DeepSeek V4.1 Flash, the 2026-10-17 release, with the same official list price and peak/off-peak schedule. They are separate IDs in separate groups, so pick one by ID; a key with only the main DeepSeek group cannot call the -sg ID.
When are DeepSeek's off-peak hours in Singapore?
Peak is Monday to Friday, 09:00–12:00 and 14:00–18:00 Beijing time. Singapore is also UTC+8, so those are the same clock times in Singapore. All other hours, including weekends, bill at half the peak rate.
Does it support tool calling, streaming and reasoning?
Yes. deepseek-v4.1-flash-1017-sg supports streaming, tool calling, thinking (reasoning_content) and automatic prefix caching, with a 1M-token context window and up to 393,216 output tokens per call.
Can I use it with Claude Code or the Responses API?
Yes. It is served over OpenAI Chat Completions, the Responses API and Anthropic Messages. For Anthropic Messages the base URL is https://api.hop-base.com (no /v1).
Can I use the same key for Claude and GPT?
Yes. A key can include up to 10 groups across platforms. Claude uses the Anthropic protocol at https://api.hop-base.com (no /v1); GPT and DeepSeek use https://api.hop-base.com/v1.
Will my data stay in Singapore?
This guide does not make data-residency commitments. If you have residency or PDPA requirements, contact an advisor before sending personal data.