Microsoft Decision-1 Released: Pricing and Model Details
Microsoft released Decision-1 on October 11 at $0.042 per million input tokens. The decision model was trained on Qwen3.5-9B, with a future rebase to MAI or OpenAI planned. Its reported price comparison lacks a named OpenAI comparator, and its release does not establish image or video generation capabilities.
Table of contents
Microsoft Decision-1 Released: Pricing and Model Details
Microsoft’s new Decision-1 model puts a specific kind of AI work in focus: choosing among defined options rather than generating a finished creative asset.
On October 11, Microsoft released Decision-1, a decision model priced at $0.042 per million input tokens. The reported pricing is approximately 58% cheaper than OpenAI, though the available account does not identify the particular OpenAI product or pricing basis used for that comparison. Decision-1 was trained on Alibaba’s Qwen3.5-9B, and Microsoft says it plans to rebase the model to MAI or OpenAI later.
Those details describe both a current release and a planned change. The Qwen3.5-9B training base is relevant to Decision-1 as announced; the MAI or OpenAI rebase is a future intention, not a feature of the released model. The quoted price is an input-token price, not a complete estimate of what any particular application will cost to run.
For people making AI images and video, the immediate question is not whether Decision-1 can render footage. It is whether a lower-priced decision model could help organize the choices around a creative workflow—and where human judgment must remain in charge.
What Decision-1 Changes—and What It Does Not
Decision-1 is news about a decision model, not an announced AI video model. The release details provided do not establish text-to-video, image-to-video, image-generation, or editing capabilities. Treating its token price as the cost of producing a video would therefore confuse two different jobs.
Its price may matter to teams that make frequent, repeatable choices. A content team might need to sort proposed concepts by campaign brief, route drafts for review, or decide which candidate deserves another round of work. Those are examples of decisions a team could define; they are not claims about Decision-1’s tested performance or supported integrations.
Microsoft’s stated plan to rebase Decision-1 is another reason to separate the model’s role from a production pipeline’s permanent rules. If a team evaluates a decision model, it should record the options it offered, the criteria used, and the outcome it accepted. That record makes it easier to tell whether a later model change affects the choices that matter.
The Price Comparison Needs Context
At $0.042 per million input tokens, Decision-1 has a clearly stated input rate. The accompanying “approximately 58% cheaper than OpenAI” figure is a reported comparison, but it is not a universal statement that Decision-1 costs less than every OpenAI model or service. Without a named comparator and the rest of the billing terms, the percentage cannot settle a full-workflow budget.
| Release detail | What it tells a production team | What remains open |
|---|---|---|
| $0.042 per million input tokens | A stated rate for input to Decision-1 | The total cost of a particular workflow |
| Approximately 58% cheaper than OpenAI | A reported price comparison | The specific comparator and whether the comparison fits a team’s use case |
| Trained on Qwen3.5-9B | The announced training base | How it will perform on a team’s own decisions |
| Planned MAI or OpenAI rebase | Microsoft intends to change the model base later | The timing and effect of that change |
That distinction is particularly important in visual production. Deciding which concept to pursue, generating an image, turning an image into video, and approving a final cut are separate activities. A low input-token rate for one of them says little about the cost or quality of the others.
Where a Decision Model Fits in AI Content Creation
Creative selection can be structured without pretending it is purely mechanical. A campaign may have several candidate scenes, each with different strengths: one communicates the product clearly, another matches the intended mood, and a third is easier to adapt across formats. A decision model could be evaluated for helping sort such options against stated criteria. A creative lead should still inspect the actual assets and make the consequential call.
The same separation helps prevent a misleading model comparison. A decision model should be judged on the choices it helps produce. An AI video model should be judged on the visual result a team can review. Neither should inherit capabilities or quality claims from the other.
A cheaper decision is useful only when it leads the team toward an asset that still meets the brief.
For teams using text to video or image to video, this suggests a practical division of labor: keep selection criteria explicit, compare the resulting visual work, and retain human review before publication. If the decision layer changes—as Microsoft says Decision-1’s model base eventually will—the team can reassess its recommendations without losing its creative standard.
Why Model Flexibility Still Matters
Decision-1 illustrates how specialized the AI model landscape is becoming. One release may target decisions and emphasize input pricing; another tool may be used to explore visual directions. A production setup benefits from keeping those roles distinct instead of expecting a single model announcement to answer every creative need.
MagicBoat gives visual teams one place to explore AI image and video creation while they assess which tools fit their work. That does not imply Decision-1 is available on MagicBoat. It means a team can keep its visual experimentation grounded in actual outputs, rather than choosing a generation workflow on the strength of a decision-model price announcement.
Keep the creative result in view. Explore image and video creation on MagicBoat, compare what you make against your brief, and reserve the final decision for the people responsible for the work.