The DeepSeek API, one key among open models
DeepSeek V3.1 Terminus is the agent-tuned release of DeepSeek's sparse MoE line: hybrid thinking mode, steadier tool calling over long chains, fewer language mix-ups. It runs here behind the same OpenAI-compatible endpoint and vl_ key as every other model in the catalog.
- 685B MoE
- Hybrid thinking mode
- Tool calling + JSON mode
- Batch at 50% off
Per token, nothing else.
Input / output per 1M tokens
$0.27 / $1.00
Batch jobs: $0.135 / $0.50 (50% off, 24h window)
openrelay/deepseek-v3.1-terminusFull catalog on the inference pricing page. Deposit $5 to get $10.
from openai import OpenAI
client = OpenAI(
base_url="https://inference.openrelay.inc/v1",
api_key="vl_••••••••", # same SDK, new base URL
)
resp = client.chat.completions.create(
model="openrelay/deepseek-v3.1-terminus",
messages=[{"role": "user", "content":
"Plan the refactor, then list the file moves as JSON."}],
)
print(resp.choices[0].message.content)The spec sheet.
What the model is, what it takes in, and the surface it serves on.
- Model id
- openrelay/deepseek-v3.1-terminus
- Architecture
- 685B sparse mixture-of-experts
- Context
- 65,536 tokens, shared prompt + completion budget
- Knowledge cutoff
- July 2025
- Endpoint
- POST /v1/chat/completions (streaming + non-streaming)
- Capabilities
- Reasoning, tool calling, JSON mode, streaming
Where DeepSeek V3.1 Terminus earns its place.
What this model is actually for, versus the rest of the catalog.
Tuned for long agent runs
Terminus is the release DeepSeek aimed at multi-step agentic work: more consistent tool calls across long code and search chains, and less language drift than V3.1. If your agents run dozens of steps, this is the difference you notice.
Reasoning without reasoning prices
The hybrid thinking mode gives chain-of-thought quality on hard problems at $0.27 in and $1.00 out per 1M tokens, a fraction of frontier proprietary reasoning rates.
No second vendor to onboard
Same base URL, same vl_ key, same billing as the rest of the catalog. Adding DeepSeek to a stack that already runs GPT-OSS or Gemma here is a one-line model swap, not a new account.
Run it in batch at half price.
Overnight DeepSeek jobs (evals, synthetic data, backfills) run through the Batch API at $0.135 in / $0.50 out per 1M tokens: JSONL up, results within 24h.
Batch Inference API overviewDeepSeek V3.1 Terminus, answered.
How do I get a DeepSeek API key?
On OpenRelay: sign up, deposit $5 (you get $10 of credit), and create a vl_ key. That one key runs DeepSeek V3.1 Terminus and every other model in the catalog through the OpenAI-compatible endpoint at inference.openrelay.inc/v1. There is no separate DeepSeek account, and the key works with the standard OpenAI SDK.
What does the DeepSeek API cost?
DeepSeek V3.1 Terminus bills at $0.27 per 1M input tokens and $1.00 per 1M output tokens here, metered per request with no subscription or minimum. Batch jobs run at half those rates.
Is there a free DeepSeek API?
Not a permanently free one anywhere serious. Here the $5 starting deposit is doubled to $10 of credit, which at these rates covers tens of millions of tokens, enough to evaluate the model thoroughly before real spend.
Is this the same model as DeepSeek's official API?
It is the open-weight DeepSeek V3.1 Terminus release, served on infrastructure OpenRelay routes and operates. You get the model's published behavior (hybrid thinking, tool calling) with US-based billing, one key across a multi-model catalog, and the option of batch pricing.
Which DeepSeek models are available?
V3.1 Terminus for chat and reasoning, and DeepSeek-OCR 2 for document transcription and extraction. Both are live; other DeepSeek ids in the catalog marked request access are discoverable but not yet runnable.
First request in five lines.
Point the OpenAI SDK at inference.openrelay.inc/v1 and run DeepSeek V3.1 Terminus per token. No contract, no minimums.