DeepSeek has just leveled up. The latest release, DeepSeek-R1-0528, is now available on Lambda’s Inference API, delivering a formidable blend of mathematical prowess, code generation finesse, and reasoning depth. This open-source model is challenging the dominance of OpenAI’s o3 and Google’s Gemini 2.5 Pro, offering you a powerful tool for complex tasks.
What’s new
DeepSeek-R1-0528 builds upon the deepseek_v3 architecture. It employs FP8 quantization, enhancing its capability to handle complex computations efficiently.
Reinforcement Learning at the Core
At the heart of R1-0528 lies a robust architecture based on the DeepSeek-V3 backbone. It employs a mixture-of-experts (MoE) model with multi-headed latent attention (MLA) and multi-token prediction (MTP), enabling efficient handling of complex reasoning tasks. This approach allows the model to learn and improve through trial and error, unlike traditional supervised learning which relies on pre-labeled data. By using RL, DeepSeek-R1-0528 can self-improve its reasoning skills to generate accurate and coherent responses.
Benchmark Performance
R1-0528 demonstrates notable improvements over its predecessor across various benchmarks.
Evaluations for models across standard benchmarks, src: Open Source Weights
Enhanced Reasoning Depth
In the AIME 2025 benchmark, DeepSeek-R1-0528 achieved an impressive 87.5% accuracy, up from 70% in its predecessor. This improvement is attributed to a deeper reasoning process, with the model averaging 23K tokens per question, nearly doubling the previous version's 12K tokens.
Improved Coding, Enhanced Inference
On LiveCodeBench, DeepSeek-R1-0528 scored 73.3%, a significant leap from its previous 63.5%. Codeforces-Div1 also saw a substantial increase, with ratings jumping from 1530 to 1930. It ranks just below OpenAI's o3 model in code generation tasks. Notably, it outperforms xAI’s Grok 3 mini and Alibaba's Qwen 3, highlighting its competitive edge in the field.
Reduced Hallucination Rate
Addressing previous concerns, R1-0528’s RL has significantly suppressed hallucination issues present in the legacy R1 version, leading to reliable and consistent outputs.
Function Calling & JSON Output Support
The model now supports JSON output and function calling, enhancing its utility in various applications.
These advancements reflect the model’s deeper reasoning, supported by its sophisticated architecture and increased token utilization.
The recent Artificial Analysis Intelligence Index further highlights DeepSeek-R1-0528’s capabilities, placing it among the top AI models available:
Artificial Analysis Intelligence Index - AI Models
The Artificial Analysis Intelligence Index ranks top AI models across seven evaluations: MMLU-Pro, GPQA Diamond, Humanity's Last Exam, LiveCodeBench, SciCode, AIME and MATH-500. DeepSeek-R1-0528 made a solid leap from 60 to 68 between January and May, edging past strong contenders like Grok and Gemini 2.5. This jump shows how the upgrade with fine-tuning boosted its reasoning and overall AI capabilities.
Getting Started with Lambda’s Inference API
DeepSeek-R1-0528's enhanced coding capabilities extend to complex tasks such as object detection. Leveraging its improved reasoning depth and support for function calling, developers can generate and integrate sophisticated computer vision solutions seamlessly.
Here's an example of how to use DeepSeek-R1-0528 via Lambda's Inference API to generate a Python script for object detection using OpenCV:
from openai import OpenAI
# Provide the endpoint to connect to Lambda's inference API
client = OpenAI(
api_key="your_api_key",
base_url="https://api.lambda.ai/v1"
)
# Choose the model and define the context
chat = client.chat.completions.create(
model="deepseek-r1-0528",
messages=[
{"role": "system", "content": "You are a computer vision expert."},
{"role": "user", "content": "Generate a Python script using OpenCV to perform object detection on a video stream, drawing bounding boxes around detected objects."}
],
)
print(chat.choices[0].message.content)
# Output Code generated by DeepSeek-R1-0528
import cv2
import numpy as np
# Initialize video capture and MobileNet SSD model
cap = cv2.VideoCapture(0)
net = cv2.dnn.readNetFromCaffe('MobileNetSSD_deploy.prototxt', 'MobileNetSSD_deploy.caffemodel')
classes = [ "traffic lights", "bicycle", "car", "truck", "motorbike", "person", "tree", "train"]
# create the bounding boxes around objects
while True:
_, frame = cap.read()
h, w = frame.shape[:2]
blob = cv2.dnn.blobFromImage(cv2.resize(frame, (300, 300)), 0.007843, (300, 300), 127.5)
net.setInput(blob)
detections = net.forward()
# assign classes to the objects
for i in range(detections.shape[2]):
confidence = detections[0, 0, i, 2]
if confidence > 0.5:
idx = int(detections[0, 0, i, 1])
box = detections[0, 0, i, 3:7] * np.array([w, h, w, h])
(x1, y1, x2, y2) = box.astype("int")
cv2.rectangle(frame, (x1, y1), (x2, y2), (0, 255, 0), 2)
cv2.putText(frame, f"{classes[idx]}: {confidence:.2f}",
(x1, y1-5), cv2.FONT_HERSHEY_SIMPLEX, 0.5, (0,255,0), 2)
cv2.imshow('Live Detection', frame)
if cv2.waitKey(1) == ord('q'):
break
cap.release()
cv2.destroyAllWindows()
Key components:
- Uses OpenCV's DNN module with MobileNet SSD (efficient for real-time)
- Processes webcam feed frame-by-frame
- Detects 20 common object categories (people, cars, traffic lights, etc.)
- Draws bounding boxes with labels and confidence scores.
Fig: Sample Output after running the object detection on video stream
This example showcases DeepSeek-R1-0528's precise, real-time object detection coding capabilities, underscoring its utility in computer vision.
Future Directions for DeepSeek
DeepSeek-R1-0528 exemplifies how reinforcement learning can revolutionize reasoning capabilities in LLMs. By minimizing dependence on supervised datasets, it sets a new standard for scalable, efficient AI development. With unparalleled performance in reasoning-intensive tasks and innovative training methodologies, R1-0528 is positioned to lead the next wave of AI advancements.
Ready to harness the power of DeepSeek-R1-0528? Integrate it into your projects via Lambda’s Inference API and embrace the future of AI development on your terms.
Want to see DeepSeek-R1-0528 in action instantly? Jump into Lambda Chat and experience its power firsthand: Try it now!