Azure AI Foundry and AOAI (Azure OpenAI Companies) retains on getting higher on a regular basis! The most recent addition in Azure AI Foundry (as of April 14, 2025, yesterday) is the GPT-4.1 mannequin, that has a beneficiant 1M context window and cutoff day for June 2024! 1 million tokens understanding “reminiscence”, which interprets to roughly 1500 pages in a doc! I’m wondering if the just lately revealed new summarizing function in Phrase makes use of this mannequin with some trick, or simply has a extremely intelligent trick on summarizing because it ought to assist summarizing paperwork as much as 3 000 pages..
What’s New: GPT-4.1 Overview
GPT-4.1 is the newest iteration of the GPT-4o mannequin, educated to excel at coding and instruction-following duties. This mannequin will enhance the standard of agentic workflows and speed up the productiveness of builders throughout all situations.
Asserting the GPT-4.1 mannequin sequence for Azure AI Foundry and GitHub builders
The GPT-4.1 context window of 1 million tokens could be very beneficiant and superior. Because the mannequin helps varied options, it is vitally usable mannequin for brokers, coding and evaluation situations. No, it’s not a reasoning mannequin, in case you are in search of evaluation focus however this one has a really massive context window that does assist to take massive variety of information into consideration. And this helps particularly with coding. If you wish to attempt it out, GPT-4.1 is already accessible for Copilot in GitHub for public preview.
OpenAI GPT-4.1 is rolling out for all Copilot Plans, together with Copilot Free. You’ll be able to entry it by way of the mannequin picker in Visible Studio Code and on github.com chat. To speed up your workflow, whether or not you’re debugging, refactoring, modernizing, testing, or simply getting began, choose “GPT-4.1 (Preview)” to start utilizing it.
OpenAI GPT-4.1 now accessible in public preview for GitHub Copilot and GitHub Fashions
What’s 1M token context good for, are brokers. As we’re transferring an increasing number of in the direction of the world the place AI has a reminiscence, and Responses/Assistants API has been implementing that already, the bigger context counts. We will proceed the dialog for an extended interval with extra info included.
Key Options of GPT-4.1
What else is included within the record? With 16K token output assist there are
Textual content, picture processing
JSON Mode
parallel operate calling
Enhanced accuracy and responsiveness
Parity with English textual content and coding duties in comparison with GPT-4 Turbo with Imaginative and prescient
Superior efficiency in non-English languages and in imaginative and prescient duties
Assist for enhancements
Assist for advanced structured outputs.
I’m very happy to see superior efficiency listed for non-English languages and naturally advanced structured outputs with JSON mode will assist massive time with brokers.
From mannequin descriptions web page, particulars are
Textual content & picture enter
Textual content output
Chat completions API
Responses API
Streaming
Operate calling
Structured outputs (chat completions)
What’s odd, is that in Be taught Max Output Tokens are 32k, and within the mannequin description (when deploying) it’s 16K.
Now that I received the mannequin deployed, it will likely be quickly the time to begin testing it.
Pricing and testing
Simply maintain within the thoughts, that the billing mannequin is completely different for as much as 128K tokens and to massive one (as much as 1M tokens) context inputs. In the mean time, once I was scripting this put up, there wasn’t any details about the pricing in Azure OpenAI Companies Pricing web page. Additionally, I can’t see GPT-4.1-mini nor GPT-4.1-nano fashions but within the catalogue. Along with the usual mannequin, you’ll get 1M context size with mini and nano as properly, however with a decrease value (on the expense of effectivity).
For extra info, try Microsoft’s announcement weblog article about GPT-4.1. From the article some further highlights:
Supervised fine-tuning for GPT-4.1 and 4.1-mini is coming quickly (this week)
Improved instruction following: The mannequin excels at following detailed directions, particularly brokers containing a number of requests. It’s extra intuitive and collaborative, making it simpler to work with for varied functions.
Enhanced coding and instruction following: The mannequin is optimized for higher dealing with of advanced technical and coding issues. It generates cleaner, less complicated front-end code, precisely identifies essential adjustments in present code, and persistently produces outputs that compile and run efficiently.
Only for a fast take a look at, I connected 4.1 to a knowledge supply and requested about “create an in depth testing plan, that solutions to all recognized dangers within the venture”. There are simply two paperwork: one a few threat administration and one with recognized dangers.
One other take a look at I did, was to ask about enhancements to this weblog draft, primarily based the plain textual content model and with a number of connected photos.
And as this weblog put up was written with a move, I didn’t use AI to generate the primary draft. Asking for higher grammar gave me loads of recommendation.
I mounted a few of these, and that hopefully helped within the readability.
These are simply easy assessments, however in time I’ll use this for extra superior situations.
Printed by
I work, weblog and discuss Future Work : AI, Microsoft 365, Copilot, Loop, Azure, and different providers & platforms within the cloud connecting digital and bodily and folks collectively.
I’ve 30 years of expertise in IT enterprise on a number of industries, domains, and roles.
View all posts by Vesa Nopanen
