DeepSeek-R1-0528 and DeepSeek-R1-0528-Qwen3-8B Now Available on Novita AI

DeepSeek has released two groundbreaking models—DeepSeek-R1-0528 and DeepSeek-R1-0528-Qwen3-8B — both now available on Novita AI, an AI cloud platform that offers developers easy AI model deployment through our straightforward API. These models deliver exceptional reasoning capabilities at competitive prices, offering significantly better performance than their predecessors.

Novita AI offers both DeepSeek-R1-0528 models at competitive rates with transparent, pay-per-use pricing:

DeepSeek-R1-0528 (Full Model):

• Input tokens: $0.7 per million tokens
• Output tokens: $2.5 per million tokens

DeepSeek-R1-0528-Qwen3-8B (Efficient 8B Model):

• Input tokens: $0.06 per million tokens
• Output tokens: $0.09 per million tokens

What is DeepSeek-R1-0528?

The DeepSeek R1 model has undergone a minor version upgrade, with the current version being DeepSeek-R1-0528. In this latest update, DeepSeek R1 has significantly improved its depth of reasoning and inference capabilities by leveraging increased computational resources and introducing algorithmic optimization mechanisms during post-training.

The model demonstrates outstanding performance across various benchmark evaluations, including mathematics, programming, and general logic. Its overall performance now approaches that of leading models, such as O3 and Gemini 2.5 Pro.

Key Improvements of DeepSeek-R1-0528

🔹 Improved benchmark performance across mathematics, coding, and reasoning tasks
🔹 Enhanced front-end capabilities for better user experience
🔹 Reduced hallucinations for more reliable outputs
🔹 Supports JSON output & function calling for seamless integration

Enhanced Reasoning Depth

Compared to the previous version, the upgraded model shows significant improvements in handling complex reasoning tasks. For instance, in the AIME 2025 test, the model’s accuracy increased from 70% in the previous version to 87.5% in the current version.

This advancement stems from enhanced thinking depth during the reasoning process: in the AIME test set, the previous model used an average of 12K tokens per question, whereas the new version averages 23K tokens per question.

Competitive Performance Against Industry Leaders

DeepSeek-R1-0528 demonstrates remarkable performance when compared to industry-leading models across multiple challenging benchmarks:

DeepSeek-R1 vs DeepSeek-R1-0528

Mathematical Reasoning Performance

Benchmark	DeepSeek R1	DeepSeek R1-0528	Improvement
AIME 2024 (Pass@1)	79.8	91.4	+11.6
AIME 2025 (Pass@1)	70.0	87.5	+17.5
HMMT 2025 (Pass@1)	41.7	79.4	+37.7
CNMO 2024 (Pass@1)	78.8	86.9	+8.1

Coding Performance Enhancement

Benchmark	DeepSeek R1	DeepSeek R1-0528	Improvement
LiveCodeBench (Pass@1)	63.5	73.3	+9.8
Codeforces-Div1 (Rating)	1530	1930	+400
SWE Verified (Resolved)	49.2	57.6	+8.4
Aider-Polyglot (Acc.)	53.3	71.6	+18.3

General Reasoning Tasks

Benchmark	DeepSeek R1	DeepSeek R1-0528	Improvement
MMLU-Redux (EM)	92.9	93.4	+0.5
MMLU-Pro (EM)	84.0	85.0	+1.0
GPQA-Diamond (Pass@1)	71.5	81.0	+9.5
SimpleQA (Correct)	30.1	27.8	-2.3
FRAMES (Acc.)	82.5	83.0	+0.5
Humanity’s Last Exam (Pass@1)	8.5	17.7	+9.2

New Tool Integration Capabilities

DeepSeek-R1-0528 introduces enhanced capabilities in function calling and tool integration:

• BFCL_v3_MultiTurn (Acc): 37.0%
• Tau-Bench (Pass@1): 53.5% (Airline) / 63.9% (Retail)

DeepSeek-R1-0528-Qwen3-8B: Compact Excellence

DeepSeek distilled DeepSeek-R1-0528’s chain-of-thought reasoning to create DeepSeek-R1-0528-Qwen3-8B from Qwen3 8B Base. This model achieves state-of-the-art (SOTA) performance among open-source models on AIME 2024, surpassing Qwen3-8B by +10.0% and matching the performance of Qwen3-235B-thinking.

This demonstrates that the chain-of-thought from DeepSeek-R1-0528 holds significant importance for both academic research on reasoning models and industrial development focused on small-scale models.

Competitive Analysis with Leading Models

Model	AIME 24	AIME 25	HMMT Feb 25	GPQA Diamond	LiveCodeBench
Qwen3-235B-A22B	85.7	81.5	62.5	71.1	66.5
Qwen3-32B	81.4	72.9	–	68.4	–
Qwen3-8B	76.0	67.3	–	62.0	–
Phi-4-Reasoning-Plus-14B	81.3	78.0	53.6	69.3	–
Gemini-2.5-Flash-Thinking-0520	82.3	72.0	64.2	82.8	62.3
o3-mini (medium)	79.6	76.7	53.3	76.8	65.9
DeepSeek-R1-0528-Qwen3-8B	86.0	76.3	61.5	61.1	60.5

Usage Improvements and Technical Details

Enhanced Usability Features

Compared to previous versions of DeepSeek-R1, the usage recommendations for DeepSeek-R1-0528 include the following improvements:

• System prompt support: Unlike earlier versions, you can now use system prompts to provide consistent context and instructions
• Automatic thinking pattern: No manual activation required—the model automatically engages its reasoning capabilities without needing “<think>\n” at the beginning of outputs

DeepSeek-R1-0528-Qwen3-8B Architecture

The model architecture of DeepSeek-R1-0528-Qwen3-8B is identical to that of Qwen3-8B, but it shares the same tokenizer configuration as DeepSeek-R1-0528. This model can run in the same manner as Qwen3-8B, but ensure all configuration files come from DeepSeek’s repository, not the original Qwen3 project.

Why Developers Choose Novita AI

Simplified AI Model Deployment

• No Infrastructure Hassle: Focus on building applications, not managing servers
• Instant Model Access: Deploy state-of-the-art models with a single API call
• Reliable Performance: Enterprise-grade infrastructure with 99.9% uptime
• Global Availability: Low-latency access from anywhere in the world

Perfect for Any Scale

• Rapid Prototyping: Test ideas quickly with immediate model access
• Production Ready: Scale seamlessly from prototype to production
• Cost Effective: Pay only for what you use with transparent pricing
• Enterprise Support: Dedicated support for high-volume applications

Ideal Use Cases

• Mathematical AI Applications: Build tutoring platforms achieving 87.5% accuracy on AIME 2025
• Code Generation Tools: Create development assistants with 73.3% success on LiveCodeBench
• Research Platforms: Deploy reasoning models that compete with industry leaders
• Educational Software: Develop step-by-step problem-solving applications

Getting Started on Novita AI

1. Sign up in seconds at Novita AI and claim your $10 free credits
2. Explpre the DeepSeek-R1-0528 (full capability) or DeepSeek-R1-0528-Qwen3-8B (efficient)
3. Get your API key from the dashboard
4. Make your first API call using our OpenAI-compatible endpoints
5. Start building with industry-leading reasoning capabilities

For python users:

from openai import OpenAI

base_url = "https://api.novita.ai/v3/openai"
api_key = "<Your API Key>"
model = "deepseek/deepseek-r1-0528"

client = OpenAI(
    base_url=base_url,
    api_key=api_key,
)

stream = True # or False
max_tokens = 1000

response_format = { "type": "text" }

chat_completion_res = client.chat.completions.create(
    model=model,
    messages=[
        
        {
            "role": "user",
            "content": "Hi there!",
        }
    ],
    stream=stream,
    extra_body={
    }
  )

if stream:
    for chunk in chat_completion_res:
        print(chunk.choices[0].delta.content or "", end="")
else:
    print(chat_completion_res.choices[0].message.content)
  
  

Conclusion

DeepSeek-R1-0528 series models deliver industry-leading performance at competitive prices—achieving 91.4% on AIME 2024 and 73.3% on LiveCodeBench while rivaling OpenAI-o3 and Gemini-2.5-Pro capabilities.

Through Novita AI’s developer-friendly platform, you get instant access to these cutting-edge models without infrastructure complexity. With transparent pricing, $10 free credits, and our intuitive API, integrating world-class reasoning into your applications has never been easier.

Novita AI is an AI cloud platform that offers developers an easy way to deploy AI models using our simple API, while also providing the affordable and reliable GPU cloud for building and scaling.