M-1: AskAIs Opens Its First-Generation Model and Generative AI Platform API

Published: June 2023

Today we are opening the API for AskAIs' first-generation model. Developers can now access two core M-1 capabilities through a unified interface: AI chat and text generation, and AI image generation.

This is not the first time M-1 has been used. The first M-1 model and AskAIs website were completed and launched in February 2023. That initial public product already offered AI chat and text-to-image generation to real users.

During the following months, we adjusted the model, website, and inference service using real feedback. In April, the team expanded development of the M1 large language model, focusing on language understanding, conversational context, content generation, and response quality. By June, we had turned the existing services into standard interfaces and opened them to external developers.

M-1 was moving from an AI website for individual users to a generative AI platform that other websites, applications, and business systems could call.

February 2023: the first M-1 model enters service

The original goal of M-1 was to build a generative AI product people could actually use. We did not want to stop at a model demonstration. We wanted the capability to work as a complete, stable, and approachable website.

AI chat

Users could ask questions, generate content, rewrite text, organize information, and seek creative suggestions in natural language. Although the conversation system was still a first-generation product, it supported everyday questions, basic multi-turn exchanges, and entry-level content creation.

AI image generation

The first website also included image generation. A user could describe a scene, subject, composition, color palette, and style, then ask the model to create an image. M-1 therefore began as a language-and-vision generation product rather than a chat model alone.

From model capability to a complete website

A model was only one part of the product. The site also needed an interface, task submission, generation status, result presentation, request management, and model scheduling.

Chat required the system to accept input, organize context, invoke the language model, and stream the result. Image generation required prompt processing, task creation, queue state, and delivery of the finished image. These seemingly basic systems are what allow a model to leave the laboratory and operate as a real product.

The first lessons from real users

Real users express themselves in more varied ways than test prompts. Some ask direct questions, others provide long backgrounds, and many want to revise the previous result. Image descriptions vary just as much across people, scenes, colors, and styles.

Usage revealed where the first generation needed work:

• multi-turn context could break;
• long instructions needed better priority detection;
• quality varied between languages;
• output consistency needed improvement;
• complex image prompts remained difficult;
• peak-time scheduling needed optimization;
• response latency could be reduced.

Feedback changed the question from “Can the model generate?” to “Can it understand consistently, stay available, and produce something genuinely useful?”

April 2023: deeper development of the M1 language model

Chat and text generation were already live in February. April was not a new beginning; it was a systematic upgrade informed by the first two months of real operation.

Instruction understanding

Users often specify subject, length, tone, format, and audience in one request. We worked on recognizing the relationship between those requirements and fulfilling them in one generation.

Multi-turn conversation

Requests such as “continue,” “rewrite,” “shorten,” or “change the tone” depend on previous context. Preserving that relationship became a core development direction.

Content generation

We expanded the experience for articles, explanations, summaries, headlines, creative work, and marketing content so M1 could complete a task, not merely answer a factual question.

Languages and inference

AskAIs served users in multiple regions, so we improved Chinese, English, and other languages as forms of expression—not just translations. At the same time, we optimized model loading, request queues, output delivery, and failure handling.

M-1 and the M1 large language model

M-1 is the complete first-generation AskAIs model and product system. It includes the website, AI chat and text generation, AI image generation, inference and scheduling, request processing, result delivery, and developer APIs.

The M1 large language model is the component responsible for language understanding, chat, and text generation. M-1 is therefore the name of the overall generative AI platform, not only one language model.

Why open an API?

The website made M-1 accessible to individual users, but developers wanted AI chat inside their own sites, content generation in their products, and text-to-image features in their applications.

An API hides complex model deployment and task infrastructure behind a standard service. Developers do not need to operate models or build a complete queue system; they submit a request and receive the generated result.

June 2023: the M-1 API opens

Chat and text generation API

Developers could send text for M1 to understand and answer. Applications included AI assistants, question answering, content tools, writing support, rewriting, summarization, embedded assistants, simple support, and creative marketing work.

Image generation API

Developers could submit text prompts and receive generated images for articles, social media, creative concepts, design drafts, personalized visuals, in-app creation, product presentation, and marketing materials.

The first API was intentionally focused, but it marked an important transition: capabilities already running on the AskAIs website were now available to other products.

From a website service to a developer platform

After the API release, M-1 was no longer only the model behind AskAIs. Individual users could keep using chat and image generation on the website, while developers could embed the same services. The website made the model visible; the API carried it into broader applications.

First-generation design principles

• Work in the real world: accept genuine user requests rather than remain an internal demo.
• Support text and images: begin as a generative platform across formats, not a chat tool alone.
• Share services between website and API: reduce capability gaps and keep upgrades aligned.
• Enable continuous iteration: let the system improve without rebuilding the platform for every model change.

Capabilities and limitations

First-generation language output could contain factual errors or misunderstand complex instructions. Professional, current, or high-stakes information still required verification. Image generation could also need several prompt revisions, especially for complex people, text, fine detail, and composition.

Context length, quality across languages, response speed, and peak-time capacity remained areas for continuing work. M-1 was not intended to solve every problem in one release; it established a product that could run, learn from feedback, and improve.

What comes next

• stronger language understanding and multi-turn context;
• longer-form generation and processing;
• better quality across languages;
• improved interpretation of complex image prompts;
• faster responses and greater service reliability;
• clearer API documentation, errors, and developer experience;
• more model capabilities and integration scenarios.

Frequently asked questions

How are M-1 and M1 different?

M-1 is the complete platform; M1 is its language-understanding, chat, and text-generation core.

When did the first website launch?

February 2023, with both chat and image generation already serving real users.

When did the API open?

June 2023, when AskAIs first opened chat, text, and image generation to external developers.

The first M-1 was only the beginning

From the model and website entering service in February, through deeper M1 development in April, to the API release in June, AskAIs completed the first stage of its transition from a model product to a developer platform.

The first generation still had much to improve, but it established the foundations that mattered most: real operation, continuous iteration, and open interfaces connecting the model to more applications.

In June 2023, the first-generation M-1 chat and image generation API officially opened.