|
|
In March, the Amazon Web Services (AWS) became the first cloud service provider to deliver Deepseek-R1 without a server launched as a fully managed, generally available model in Amazon Bedrock. Since then, customers have used Deepseek-R1 capabilities via Amazon Bedrock to create generative AI applications using the robust backing railing and comprehensive tools for safe AI deployment.
Today I am excited to announce Deepseek-V3.1 now available as a fully managed endowment model in Amazon Bedrock. Deepseek-V3.1 is a hybrid model of open weight that switches between thinking mode (sophisticated thinking) for detailed analysis step by step and mode not to think (direct response) to faster answers.
According to Deepseek, the Deepseek-V3.1 thinking regime achieves comparable response quality with better results, a stronger multi-stage justification for comprehensive search tasks and high gains in thinking efficiency compared to Deepseek-R1-0528.
| Benchmarks | Deepseek-v3.1 | Deepseek-R1-0528 |
|---|---|---|
| Browser | 30.0 | 8.9 |
| Wowsecomp_zh | 49.2 | 35.7 |
| Bell | 29.8 | 24.8 |
| xbench-a-beepsearch | 71.2 | 55.0 |
| Frames | 83.7 | 82.0 |
| Simpleqa | 93.4 | 92.3 |
| Seal0 | 42.6 | 29.7 |
| SWE-BENCH Verified | 66.0 | 44.6 |
| SWE-BENCH Multilingual | 54.5 | 30.5 |
| Terminal bench | 31.3 | 5.7 |
The performance of the Deepseek-V3.1 in the use of agents’ instruments and tasks has improved significantly through optimization after training compared to previous Deepseek models. Deepseek-V3.1 also supports more than 100 languages with almost native knowledge, including significantly improved low-source languages that lack large monolingual or parallel corpuses. You can create global applications to provide increased accuracy and reduced hallucinations compared to previous Deepseek models, retaining their visibility in their decision -making process.
Here are your key cases of use using this model:
- Generation -Deepseek-v3.1 excels in coding tasks with improvement in software engineering benchmarks and code agents’ capabilities, ideal for automated code generation, debugging and working procedures of software engineering. It works well on coding scales and at the same time effectively provides high -quality results.
- Agriculture Ai -Model contains improved tools calling optimization after training, so it is strong when using tools and agent workflows. It supports structured calls to tools, code agents and search agents and place it as a solid choice for building AI autonomous systems.
- Business applications – Deepseek models are integrated into various chat platforms and productivity tools, increasing user interactions, and support customer service workflows. Multilingual abilities and cultural sensitivity of the model are suitable for global business applications.
As I have already mentioned in my previous post, when implementing publicly available models, it will carefully consider the requirements for the protection of personal data when implementing in your production environment, check the output and monitor your data security results, AI and model evaluation.
You can access Amazon Bedrock’s business safety features and implement warranty adapted to your applications and AI -responsible AIs with Amazon Bedrock Guardrails. You can also evaluate and compare models and identify the optimal model for your cases using the Amazon Bedrock tools.
Start with Deepseek-V3.1 in Amazon Bedrock
To test the Deepseek-V3.1 in the Amazon Bedrock console, choose Chat/text under Field In the left pane of the offer. Then choose Select the model at the top left and select Deepseek as a category of a Deepseek-v3.1 as a model. Then choose Apply.
By means of a selected Deepseek-v3.1 Model, I run the following quick example about the decision of technical architecture.
Outline the high-level architecture for a scalable URL shortener service like bit.ly. Discuss key components like API design, database choice (SQL vs. NoSQL), how the redirect mechanism works, and how you would generate unique short codes.
You can switch on and off Justification Mode for generating the chain of thought of response before the final conclusion.
You can also access the model using the AWS (AWS CLI) and AWS SDK command line. This model supports how InvokeModel and Converse API. You can see a wide range of code examples for multiple cases and a number of programming languages.
If you want to know more, visit the inference parameters and answers Deepseek Model in AWS documentation.
Now available
Deepseek-V3.1 is now available on the American West (Oregon), Asia Pacific (Tokyo), Asia Pacific (Mumbai), Europe (London) and Europe (Stockholm) AWS. For future updates, see the entire Region list. You want to learn more, check out the Amazon Bedrock page and Amazon Bedrock Pricking.
Try Deepseek-V3.1 Try today in the Amazon Bedrock Console and send feedback to AWS Re: Post for Amazon Bedrock or through the usual AWS support contacts.
– Channy
Updated September 19, 2025 – Removes the section of access to the model. Amazon Bedrock will simplify access to all server -free foundation models and all new models by automatically allowing them to enable them for each AWS account, eliminating the need to manually activate access through the Bedrock console. The access site will be retired on October 8, 2025 Account administrators retain full control over the model access through AWS IAM and service control principles to limit the model’s access as needed.