Microsoft Launches Decision-1 AI Model, 35x Faster and Up to 200x Cheaper Than GPT-6 Sol
Microsoft has introduced Microsoft-Decision-1, a specialised artificial intelligence model designed to make decisions between predefined options rather than generate long-form text. The company says the model can deliver substantially faster results and lower operating costs for specific business and AI automation tasks.
According to Microsoft’s published evaluations, Decision-1 was up to 35 times faster than OpenAI’s GPT-6 Sol in a benchmark comparison. Microsoft also reported that, in an internal Xbox Research workload, the model delivered comparable quality at more than 14 times the speed and up to 200 times lower cost.
The model targets developers and businesses that need AI systems to classify information, rank choices, verify results, and control automated workflows.
What Is Microsoft-Decision-1?
Microsoft-Decision-1 is a decision-focused AI model. Instead of generating paragraphs of text like a conventional chatbot, it evaluates a question against a predefined set of possible answers and assigns a probability score to each option.
For example, a business could use the model to classify a customer complaint, determine which department should handle a support request, or assess whether an AI-generated response meets a particular quality standard.
The model supports several types of structured decisions, including:
- Yes-or-no questions
- Multiple-choice selection
- Ratings and rankings
- Classification and routing
- Verification of AI-generated responses
- Evaluation of actions taken by AI agents
This focused design could make Decision-1 useful in applications where speed, consistency, and processing costs matter more than the ability to write detailed explanations.
Microsoft-Decision-1 Specifications
Microsoft has designed the model for developers who want to integrate decision-making capabilities into existing applications and AI-powered workflows.
| Specification | Details |
|---|---|
| Model name | Microsoft-Decision-1 |
| Developer | Microsoft |
| Base model | Alibaba’s Qwen3.5-9B |
| Main purpose | Structured decision-making |
| Input type | Text |
| Context window | Approximately 32,000 tokens |
| Output | Probability scores for predefined options |
| Reported input price | $0.042 per million tokens |
| Output token price | Free, according to Microsoft’s published pricing |
| Availability | Microsoft Foundry |
Microsoft post-trained the underlying Qwen model to handle decision-scoring tasks. The company has also indicated that it plans to adapt the approach to other model families, including its own MAI models and OpenAI models.
Microsoft Says Its AI Model Is 35 Times Faster Than GPT-6 Sol
Speed is one of Microsoft’s main selling points for Decision-1.
In its own evaluation across 36 benchmarks covering nearly 150,000 questions, Microsoft reported that Decision-1 achieved the highest average accuracy among the models included in the comparison.
The company also reported that Decision-1 was 2.5 times faster than the runner-up in its benchmark comparison and approximately 35 times faster than GPT-6 Sol.
These results concern specific decision-making tasks. They should not be interpreted as proof that Decision-1 is 35 times faster or more capable than GPT-6 Sol for every AI application.
The distinction matters because the two models serve different purposes. A general-purpose model may be needed to write reports, explain complex subjects, or generate code, while a specialised decision model can be more efficient when the required output is simply a selection or score.
Reported Performance Comparison
| Metric | Microsoft’s Reported Result |
|---|---|
| Benchmarks evaluated | 36 |
| Questions covered | Nearly 150,000 |
| Average benchmark accuracy | Highest among models included in Microsoft’s comparison |
| Speed advantage over GPT-6 Sol | Approximately 35 times in the reported comparison |
| Speed advantage over the runner-up | Approximately 2.5 times |
These figures come from Microsoft’s evaluation. Results may differ under independent testing or in real-world applications with different workloads.
Xbox Research Reports Up to 200 Times Lower Costs
One of the most notable claims concerns the cost of processing large volumes of customer feedback.
Microsoft says Xbox Research used Decision-1 to categorise more than 10,000 pieces of player feedback and reviews into predefined themes. The team reported that the model delivered quality comparable to GPT-6 Sol for this task while running more than 14 times faster and costing up to 200 times less.
This example highlights the potential value of smaller, specialised models in business operations.
Companies regularly process large quantities of customer reviews, support tickets, survey responses, and product feedback. Many of these tasks require accurate classification rather than lengthy written answers.
If a specialised model can perform such work at a significantly lower cost, businesses could potentially automate more of their routine processing without increasing their AI budgets at the same rate.
However, the reported savings relate to Microsoft’s particular workload and testing conditions. Actual savings will depend on the application, model usage, and implementation costs.
Microsoft Also Tests Decision-1 With Copilot
Microsoft has also evaluated Decision-1 within its Copilot work.
According to the company’s reported internal testing, Decision-1 delivered decision quality comparable to GPT-5.6 Luna while running around 100 times faster in the relevant workload.
This suggests that decision-focused AI could serve as a supporting component within larger AI systems.
Rather than asking one large model to handle every step, an application could use a specialised model to classify a request, verify an output, or select the next action. A general-purpose model could then handle tasks that require more extensive reasoning or text generation.
Such an approach could help developers reduce delays and costs in applications that make multiple AI decisions in sequence.
How Businesses Could Use Microsoft’s New AI Model
Decision-1 could be particularly useful in systems that repeatedly make straightforward, structured decisions.
1. Customer Support
Businesses could classify incoming support requests and direct them to the correct department or automated workflow.
2. E-Commerce
Online retailers could use decision models to categorise product feedback, prioritise customer issues, and organise incoming requests.
3. AI Agents
AI agents often need to decide whether to continue a task, retry an action, use another tool, or request human assistance. A specialised decision model could help manage these steps.
4. Quality Control
Organisations could evaluate AI-generated responses against predefined criteria before using them in a business process.
5. Content Moderation and Classification
The model could help categorise text according to established rules, although high-impact or sensitive decisions would still require appropriate safeguards and oversight.
How Much Does Microsoft-Decision-1 Cost?
Microsoft lists an input price of $0.042 per million tokens, with output tokens priced at zero under its published pricing.
| Token Type | Reported Price per Million Tokens |
|---|---|
| Input | $0.042 |
| Output | Free |
At this rate, processing 100 million input tokens would cost approximately $4.20 in input-token charges, before any applicable infrastructure, integration, or other service costs.
The low token price reflects the model’s specialised role. It primarily scores a set of predefined choices instead of producing lengthy responses.
Businesses should still compare the total cost of deployment, including API access, application development, monitoring, and any additional models needed to complete a workflow.
What About Accuracy and Safety?
Microsoft says it tested Decision-1 across a range of benchmarks covering decision quality and safety. Its published evaluations included tests involving harmful content, jailbreak attempts, and prompt injection.
The company also reported that the model changed its decision an average of 1.3% of the time when equivalent requests were reworded or their options reordered.
Consistency is important in automated systems because minor wording changes should not unnecessarily alter the outcome of a routine decision.
However, benchmark results do not guarantee that a model will behave correctly in every real-world situation. Businesses should test it against their own data, monitor its performance, and establish procedures for uncertain or incorrect decisions.
Microsoft also cautions against relying on the model as the sole basis for consequential decisions affecting people, including decisions involving employment or credit.
Is Decision-1 a Replacement for GPT-6 Sol?
Not for every task.
Microsoft-Decision-1 and GPT-6 Sol have different strengths. Decision-1 is designed to select between predefined options quickly and economically, while a general-purpose model is better suited to tasks requiring open-ended explanations, detailed writing, or broader problem-solving.
A business that needs to write a report or develop a complex application may still require a general-purpose model. However, a company processing thousands of support tickets or checking structured outputs may benefit from a specialised decision model.
The two approaches could therefore complement each other rather than compete directly.
The Bigger Picture: Smaller AI Models for Specific Tasks
The launch of Decision-1 reflects a broader shift towards using specialised AI models alongside large, general-purpose systems.
As businesses deploy AI agents and automated workflows, the cost of making repeated decisions can become significant. Even small delays can accumulate when a system performs many steps in sequence.
A specialised model that handles routine decisions efficiently could reduce both latency and operating expenses, leaving larger models to focus on tasks that genuinely require more extensive reasoning or content generation.
For developers, the important question is not simply which AI model is the biggest or most powerful. It is which model delivers the best combination of accuracy, speed, reliability, and cost for a particular job.
Microsoft-Decision-1 offers one approach to that challenge. Its reported performance is promising for structured decision-making, although businesses should validate the results independently before relying on it for critical operations.
Source note: The performance, pricing, and internal-use figures in this report are attributed to Microsoft’s published information and should be understood within the specific testing conditions described by the company.