We picked the DeepSeek R1 API because it shows its reasoning — our researchers check the steps, not just the answer.
Loading model information...
Input
DeepSeek R1 (0528)
Hi! I'm a helpful AI assistant. What can I do for you?
DeepSeek R1 API — The Open Reasoner That Shows Its Work
The DeepSeek R1 API runs the full open-weights DeepSeek-R1 — a reasoning model trained with reinforcement learning to think step by step across math, code, and logic, and to show that thinking.
Most models hand you an answer; the DeepSeek R1 API hands you the thinking too. DeepSeek-R1 is a full open-weights reasoning model trained with large-scale reinforcement learning, and it streams a visible chain-of-thought before its conclusion — so you can follow, audit, or gate on the derivation instead of trusting a black box. The 0528 refresh sharpened that reasoning, cut hallucination, and added JSON output and function calling, making the DeepSeek R1 API steadier on long, multi-step problems.
You reach the DeepSeek R1 API through RouterBase on the OpenAI chat-completions protocol — set the model to deepseek-r1-0528 and your current client just works. The DeepSeek R1 API bills $0.55 per 1M input tokens, $2.19 per 1M output, and $0.14 per 1M cached reads, all 15% under list, on one key shared with the whole catalog.
What sets the DeepSeek R1 API apart
A frontier reasoner you can read, audit, and self-host.
Shows its work
The DeepSeek R1 API streams a visible chain-of-thought before the answer, so every conclusion arrives with the steps that led there.
Reinforcement-trained depth
Large-scale RL gives the DeepSeek R1 API genuine multi-step reasoning on math, code, and logic — not pattern-matched shortcuts.
Sharper since 0528
The 0528 update deepened reasoning and cut hallucination, so the DeepSeek R1 API holds its line over long derivations.
Structured output
JSON mode and function calling return in the standard OpenAI shape, so the DeepSeek R1 API slots into agents and pipelines.
Open weights
DeepSeek-R1 is open-weight — evaluate it anywhere, then let the DeepSeek R1 API host it at $0.55 / $2.19 per 1M, 15% below list, no GPUs of your own.
OpenAI-shaped
One chat-completions endpoint, one RouterBase key — point at routerbase.com/v1, name the model, and skip the DeepSeek SDK entirely.
Wire it up
Your first DeepSeek R1 API call, in three steps
From key to chain-of-thought in a few minutes.
Create a key
A single RouterBase key reaches the DeepSeek R1 API and everything else in the catalog.
Point and name
Send any OpenAI client to routerbase.com/v1 with model=deepseek/deepseek-r1-0528. Streaming and the DeepSeek R1 API reasoning trace are on by default.
Follow the thinking
Read the chain-of-thought beside the answer; usage returns per response so cost and reasoning length stay visible.
Pay only for what you use
RouterBase passes through partner-tier pricing. Compared against the model's official published API rate.
Teams building on the DeepSeek R1 API
Reasoning you can read, running in production.
The 0528 update was the turning point; the DeepSeek R1 API stopped drifting on our longest derivations.
Open weights let us evaluate offline, then the DeepSeek R1 API let us ship the same model without a GPU fleet.
JSON mode plus visible reasoning from the DeepSeek R1 API means we get structured answers we can actually trust.
For hard math the DeepSeek R1 API is our default — the chain-of-thought is usually correct enough to keep.
Function calling on the DeepSeek R1 API put a real reasoner inside our agent loop.
We audit every high-stakes decision by reading the DeepSeek R1 API trace; no other endpoint gives us that.
At $0.55 in and $2.19 out, the DeepSeek R1 API finally fit frontier-grade open reasoning into our budget.
One key, one protocol — we swapped to the DeepSeek R1 API and our old OpenAI code never noticed.
DeepSeek R1 API — questions worth asking
The essentials before you build on it.
It is RouterBase's hosted access to DeepSeek-R1 (0528) — a full open-weights reasoning model with a transparent chain-of-thought, served over an OpenAI-compatible endpoint.