DeepSeek-R1-0528 is a fixed-version upgrade of DeepSeek R1, focused on improving reasoning depth for complex problems. It is suitable for math problem solving, programming, and multi-condition logical analysis. It supports system prompts and enhances function-calling capabilities.
Point the Base URL to api.acedata.cloud, configure your platform API key and the model ID below, and use your compatible SDK or client.
api.acedata.clouddeepseek-r1-0528import os
from openai import OpenAI
client = OpenAI(
api_key=os.environ["ACEDATACLOUD_API_KEY"],
base_url="https://api.acedata.cloud/v1",
)
response = client.chat.completions.create(
model="deepseek-r1-0528",
messages=[{"role": "user", "content": "Hello!"}],
)
print(response.choices[0].message.content)Choose an available protocol for this model. OpenAI SDK uses a Base URL ending in /v1; Anthropic SDK uses the root URL. See each guide for protocol-specific parameters, tools and response formats.
First, clarify this model's input and standard calling method.
Native model characteristics are for model selection; this platform's input limits, available parameters, and billing are subject to this model's API and pricing. Continuous output for Chat Completions uses stream, and the client is responsible for saving message history.
Learn what deepseek-r1-0528 can bring to your work.
The core improvement in 0528 is deeper reasoning, making it suitable for problems involving multiple conditions, branches, and intermediate conclusions. You can ask it to first organize the known conditions, then provide a solution, key derivations, and verification. Its official AIME 2025 score improved from 70.0 for the original R1 to 87.5, but individual problem results should still be checked.
For programming tasks, you can provide requirements, existing code, error messages, and test conditions at the same time, allowing the model to analyze the cause and propose changes. Its official LiveCodeBench (2408-2505) score improved from 63.5 to 73.3. It is suitable for generating candidate implementations and testing ideas, rather than treating generated code as a verified deliverable.
The official recommendation for 0528 supports system prompts and removes the requirement to forcibly add a reasoning prefix. You can standardize language, task boundaries, and output structure so every turn follows the same rules, then validate answers with real tests.
Start with specific tasks to find where the model can be effective.
Keep the version of materials submitted to deepseek-r1-0528 and the actual responses, distinguishing original facts, model suggestions, and actions already completed by the application. Before structured results enter the system, check required fields, value types, and business rules to avoid turning missing information directly into definitive records.
Provide input-output specifications, boundary conditions, code snippets, and error logs so the model can compare algorithm approaches, explain failures, and generate modification suggestions. Deliverables may include candidate code, complexity explanations, and a test checklist; then bring actual execution results back into the conversation to further narrow the problem scope.
Organize objectives, hard constraints, available options, and evaluation criteria into text, ask the model to compare them item by item, and explain which assumptions the conclusions depend on. Suitable for technical solution reviews and rule-based reasoning, delivering comparison tables, recommendation rationales, and items to be confirmed; when key conditions are missing, supplement them first before forming conclusions.
Choose based on task complexity, input materials, and expected results.
If tasks focus on complex mathematics, programming, and multi-step logic, 0528 is more worthwhile as an upgrade candidate, as its official related benchmarks are significantly better than the original R1. However, do not equate reasoning improvements with greater accuracy for all questions and answers. It is recommended to use real problems to compare correctness, answer length, and call consumption; simple rewriting may not require deep reasoning.
Use /v1/chat/completions and explicitly set model=deepseek-r1-0528. Existing OpenAI-compatible applications can retain their message and result handling; when integrating, configure the platform address, API Key, and exact model name.
Arrange the inputs first, then connect them to the corresponding application workflow.
Provide input size, memory limits, time requirements, and typical counterexamples, and specify the programming language to use.
Explicitly select deepseek-r1-0528 in the Chat Completions request, and organize the context, materials, and output requirements for this request into messages. First use a task with a clearly defined scope to check the response, then include actual review or test feedback in the next round of messages.
Design tasks directly from the following inputs and acceptance priorities.
First compare the complexity of two algorithms, then provide implementation and tests: include empty input, duplicate elements, and extreme scale; explain the conditions protected by each test.
Run actual tests and performance examples, and check whether the complexity argument is consistent with the implementation; compare with the original R1 using the same problem, and do not use benchmark scores as a substitute for acceptance.
Before formal use, understand the output quality and scope of capabilities.
Answers to common questions when using deepseek-r1-0528.
The focus is on reasoning depth and performance on complex tasks, with improved mathematics, programming, and general logic capabilities, along with support for system prompts and improved function calling. It is not a completely different product line; whether it is worth replacing the original R1 should be determined based on actual tasks rather than evaluation scores alone.
No. The official usage guidance for 0528 no longer requires forcibly adding a thinking prefix. A more practical approach is to clearly provide the task, conditions, and delivery requirements, and request key derivations and verification. When using system messages, you can consistently specify the language and response structure.
Submit model=deepseek-r1-0528 and messages to /v1/chat/completions. Read normal results from choices[].message.content; for streaming calls, obtain incremental results through stream. Use this platform's API Key, and set the full base URL according to the SDK you use.
Generating code and executing code are two different things. R1-0528 can analyze problems and propose candidate implementations, but an ordinary text response does not mean the code has been run. To integrate a code execution tool, the selected entry point must support tool interaction for this model and have an available execution environment and permissions configured; whether it runs successfully should be determined by the tool's returned results. It is recommended to first run tests in an isolated environment, then submit logs and failure examples to the model for further troubleshooting. Do not directly execute unreviewed code.
No. DeepSeek-R1-0528-Qwen3-8B is an independent variant distilled from Qwen3 8B Base using the chain of thought of R1-0528. The deepseek-r1-0528 on this page refers to the 0528 version of R1; their evaluations or deployment requirements must not be mixed.
Choose a development approach or application suited to this model and follow its setup guide.
↗OpenAI Chat Completions API
SDK installation, authentication and examples
↗Using Ace Data Cloud with Cherry Studio
Chat client setup · Base URL, API key and model
↗Using Ace Data Cloud with ChatBox
Chat client setup · Base URL, API key and modelInput
$0.36per 1M tokens
Official $0.55
Output
$1.44per 1M tokens
Official $2.19
You pay$13,584for 142,800 credits · $0.0951 / credits46% OFF
You're billed in credits. Pick a top-up tier to see its per-credit rate — the larger the top-up, the cheaper each credit.
This is only a basic call example. See the full docs for more parameters and advanced usage.
View Docs