The MAI-Thinking-1 prompt generator gives you 20 free, copy-ready prompts for Microsoft's first in-house reasoning model. Complex analysis, multi-step reasoning, code architecture, legal review — structured for MAI-Thinking-1's 256K context window. Announced at Build 2026.
The MAI-Thinking-1 prompt generator on this page provides 20 free, professionally crafted prompts for MAI-Thinking-1 — Microsoft's first fully in-house reasoning model, announced at Microsoft Build 2026 on June 2, 2026. MAI-Thinking-1 was trained from scratch on enterprise-grade, commercially licensed data with zero distillation from OpenAI models, using a sparse Mixture-of-Experts architecture with 35 billion active parameters and a 256,000-token context window.
Unlike general-purpose frontier models, MAI-Thinking-1 is a reasoning model — it works through complex problems step by step before producing a conclusion. This makes it dramatically more effective for high-stakes structured tasks: legal analysis, financial modeling, scientific reasoning, software architecture review, and long-context document synthesis. Microsoft reports that independent raters prefer it over Claude Sonnet 4.6 in blind tests.
Every prompt below is structured specifically for MAI-Thinking-1's reasoning strengths — with role assignment, explicit step-by-step instructions, and reasoning anchors that activate the model's chain-of-thought capability. Copy any prompt and paste directly into Microsoft Foundry, Azure AI Foundry, or any compatible platform.
Reasoning models respond differently from general-purpose models. Use this framework for maximum output quality:
Click any prompt to copy — paste into Microsoft Foundry, Azure AI Foundry, or any MAI-Thinking-1 compatible endpoint
You are a senior contract attorney. A SaaS company has provided the following Master Service Agreement clause: [paste clause]. Analyze it step by step. First, identify every obligation it places on the customer. Second, identify every obligation it places on the vendor. Third, flag any asymmetric liability risks for the customer. Fourth, propose specific redline language to balance those risks. Show your reasoning at each step before giving a final recommendation.
Our API latency spiked from an average of 120ms to 2,400ms at 14:32 UTC on June 1, 2026. The spike lasted 18 minutes, affected only our /checkout endpoint, and coincided with a deployment of build #4421. Database query times remained under 40ms throughout. Walk me through a structured root-cause analysis: (1) list the most plausible hypotheses ranked by likelihood, (2) describe the diagnostic evidence that would confirm or eliminate each, (3) propose the most probable cause given what's known, (4) suggest an immediate mitigation and a permanent fix.
I am studying the relationship between chronic sleep fragmentation and insulin resistance in adults aged 40–60. My dataset shows a correlation of r=0.61 (p<0.001) between sleep efficiency below 80% and HOMA-IR scores above 2.5. Generate three distinct mechanistic hypotheses that could explain this relationship. For each hypothesis: (a) describe the biological pathway, (b) identify what additional data would support it, (c) identify what data would falsify it. Then rank the three hypotheses by explanatory power and scientific precedent.
Conduct a structured competitive analysis for a B2B project management SaaS entering a market dominated by Asana, Monday.com, and Linear. Think through this in four phases: (1) Identify the dimensions on which these competitors most differentiate from each other. (2) Map each competitor's primary customer segment and willingness-to-pay range. (3) Identify the whitespace — the underserved customer segment or capability gap none of them address well. (4) Recommend a positioning strategy that would be defensible in year one and scalable by year three. Show your reasoning at each phase.
Review the following Python microservice architecture and reason through its failure modes: [paste architecture diagram or description]. For each component, reason about: (1) single points of failure, (2) bottlenecks under 10× load, (3) data consistency risks if a component crashes mid-operation. Then produce a prioritized list of architectural improvements, ordered by impact-to-effort ratio. For the top three changes, write the specific code or config change required.
A Series B SaaS startup has the following metrics: ARR $8.2M growing 110% YoY, net revenue retention 118%, CAC $4,200, LTV $31,000, gross margin 74%, burn rate $620K/month, 14 months runway. Reason through a DCF-adjacent valuation step by step: (1) normalize the growth trajectory to a conservative, base, and optimistic scenario, (2) estimate a terminal multiple for each scenario using comparable public SaaS comps, (3) discount back at an appropriate risk-adjusted rate, (4) provide a valuation range with explicit assumptions for each bound. Show all reasoning.
A hospital AI triage system has a documented performance gap: it correctly prioritizes White patients 91% of the time but Black patients only 78% of the time, due to a proxy variable (insurance type) that correlates with race in the training data. The hospital has two options: (A) disable the AI and return to manual triage, accepting slower processing and more human variability; (B) keep the AI running while a retraining project takes 6 months. Reason through this decision using three ethical frameworks — utilitarian, deontological, and justice-based. Then synthesize a recommendation that accounts for all three perspectives.
The following Python function is producing unexpected output. Input: [paste code]. Expected output: [describe expected]. Actual output: [paste actual]. Reason through the bug step by step: (1) trace the execution path for the given input, (2) identify the exact line where the state diverges from expectation, (3) explain WHY that divergence happens (not just what it is), (4) propose a fix, (5) write a test case that would catch this class of bug in the future.
A consumer electronics company sources 60% of its NAND flash from a single Taiwanese supplier, 30% from a Korean supplier, and 10% from spot market. A geopolitical risk model suggests a 15% annual probability of a 6-month supply disruption from Taiwan. Walk through a structured risk quantification: (1) model the expected annual cost of a disruption at current inventory levels, (2) calculate the NPV of adding a third supplier at 20% premium but half the disruption probability, (3) recommend the optimal sourcing split that minimizes total expected cost including diversification premium. Show all assumptions explicitly.
I have attached three quarterly earnings call transcripts from a public company (Q2, Q3, Q4 2025). Synthesize them as follows: (1) identify the three strategic themes management repeated across all three calls, (2) identify where management's forward guidance in Q2 and Q3 proved incorrect by Q4, and explain what that implies about forecast reliability, (3) extract the five most specific numerical commitments made, note which were met or missed, (4) produce a one-paragraph investor takeaway that a fund manager could use in a memo.
Prove, step by step with full justification at each step, that the sum of the first n odd positive integers equals n². Use mathematical induction. At each step: (1) state what you are proving, (2) complete the proof step with full algebraic working, (3) explain in plain English why that step is valid. After the formal proof, explain the geometric intuition — why does this identity have a visual interpretation, and what is it?
The city of Austin, TX is considering a 15% vacancy tax on commercial properties that have been empty for more than 12 months. Reason through the second and third-order effects: (1) immediate first-order effects on landlord behavior, (2) second-order effects on the commercial real estate market, (3) third-order effects on small business availability and city tax revenue, (4) likely unintended consequences, (5) a recommendation for whether to pass the policy as written, with specific suggested amendments to address the top two unintended consequences.
Design a distributed task queue for a fintech company processing 50,000 payment events per second with at-most-once delivery semantics and sub-100ms p99 latency. Reason through the design in this order: (1) partition strategy — how do you shard by customer ID without hotspots, (2) durability guarantees — what happens to in-flight messages during a broker failure, (3) consumer group coordination — how do you handle consumer lag without reprocessing, (4) observability — the three most critical metrics to alert on and why. Then draw the component diagram in ASCII and label each edge with the protocol used.
I believe that universal basic income would reduce labor force participation and harm long-term economic growth. Steelman the strongest possible counterargument to this position. First, identify the most compelling empirical evidence against my view. Second, identify the strongest theoretical mechanism that would explain why my concern is wrong. Third, identify the conditions under which my concern would be valid and those under which the counterargument would hold. Do not give me a balanced 'on one hand / on the other hand' response — give me the most intellectually honest challenge to my stated position.
A mobile app has the following metrics: Day 1 retention 48%, Day 7 retention 19%, Day 30 retention 8%. The median session length is 4.2 minutes. The core action (completing a workout) is completed by 31% of D1 users. The push notification opt-in rate is 71%. Reason through a diagnosis: (1) identify which metric is the leading indicator of the retention problem, (2) propose the top three interventions ranked by expected impact, (3) for each intervention, describe the experiment design (hypothesis, primary metric, guardrail metric, sample size rationale, run time). Show your reasoning for each ranking decision.
A 52-year-old male presents with: fatigue for 6 weeks, unintentional weight loss of 4kg, night sweats, and a palpable left supraclavicular lymph node. No fever. Normal CBC except mild normocytic anemia (Hgb 11.2). Reason through a differential diagnosis step by step: (1) list the top five diagnoses in order of probability given this presentation, (2) for each, identify the one test that would most efficiently confirm or exclude it, (3) identify the diagnosis that would be most dangerous to miss, (4) recommend a sequenced diagnostic workup that minimizes time to diagnosis for the most dangerous possibilities first. [This is for educational/training purposes only — not medical advice.]
I am negotiating a software licensing deal where I am the vendor. The buyer is a Fortune 500 company, they have three competing bids, and my price is 18% higher than the next competitor. Their stated priority is implementation speed. My advantages: best-in-class SLA guarantees, dedicated implementation team, and a client referral from their direct competitor. Reason through a negotiation strategy: (1) what is their BATNA and how does it constrain their leverage, (2) what is my ZOPA (zone of possible agreement), (3) what concessions can I make that cost me little but are high value to them, (4) how do I reframe the price gap in terms of TCO rather than sticker price, (5) write the opening line I should use in the negotiation call.
Reason through this counterfactual: if the Allied forces had not successfully conducted Operation Overlord (D-Day) in June 1944, what is the most plausible alternative path to the end of World War II in Europe? Structure your reasoning: (1) what conditions would have changed in mid-1944, (2) what would the Eastern Front trajectory have been without the Western Allied pressure, (3) what would Germany's resource and morale situation have been by late 1944, (4) identify the key branching points where outcomes could have diverged most dramatically. Distinguish clearly between well-evidenced reasoning and speculative inference.
A US-based startup is expanding to the EU and must comply with GDPR, the EU AI Act (if applicable), and NIS2. Their product is a B2B SaaS HR analytics platform that processes employee performance data for enterprise clients. Reason through their compliance gaps: (1) classify their AI system under the EU AI Act risk tiers and explain the classification rationale, (2) identify the three highest-risk GDPR gaps for processing employee data as a data processor, (3) assess NIS2 applicability and, if applicable, the top three obligations they are likely missing, (4) prioritize these gaps by severity and produce a 90-day remediation roadmap.
A B2B startup has been operating for 22 months with $1.8M ARR but growth has stalled at 15% YoY for the last two quarters. Their original thesis was 'AI-powered contract review for mid-market legal teams.' They have 47 paying customers but only 8 have expanded their contract. NPS is 31. The founding team has identified three pivot options: (A) narrow to a vertical (construction contracts only), (B) broaden to all document review, (C) reposition as a compliance monitoring platform. Reason through the pivot decision: (1) diagnose why growth stalled using the data given, (2) evaluate each option against the core customer evidence, (3) identify the one assumption in each pivot that must be true for it to work, (4) recommend a structured experiment to test the most promising pivot within 60 days without burning runway.
MAI-Thinking-1 is a specialized reasoning model — compare it against both reasoning-focused and general-purpose frontier alternatives:
| Model | Type | Context Window | Access | Best For |
|---|---|---|---|---|
| MAI-Thinking-1 ★ (Microsoft) | Reasoning | 256K tokens | Private preview (Foundry) | Multi-step analysis, legal, financial, code review |
| GPT-5.5 (OpenAI) | General + Reasoning | 128K tokens | ChatGPT (default) | Broad tasks, creative, instruction-following |
| Gemini 4 (Google) | Multimodal | 1M tokens | Gemini Advanced | Multimodal tasks, ultra-long context, agentic |
| Gemini 3.5 Flash (Google) | Fast General | 128K tokens | Gemini app (free) | Speed, everyday tasks, agentic workflows |
| MAI-Code-1-Flash (Microsoft) | Coding | 64K tokens | GitHub Copilot / VS Code | Code completion, inline suggestions, efficiency |
★ MAI-Thinking-1 announced June 2, 2026 at Microsoft Build 2026. Available in private preview via Microsoft Foundry. General availability rollout for Azure enterprise customers in progress.
The MAI-Thinking-1 prompt generator on this page provides 20 free, copy-paste prompts for Microsoft's MAI-Thinking-1 reasoning model — announced at Microsoft Build 2026 on June 2, 2026. MAI-Thinking-1 is Microsoft's first in-house reasoning model trained entirely without OpenAI data. Every prompt on this page is structured to leverage MAI-Thinking-1's specific strengths: complex multi-step reasoning, long-context analysis, and structured problem decomposition. Copy any prompt and paste it into Microsoft Foundry, Azure AI Foundry, or any platform that has deployed MAI-Thinking-1.
MAI-Thinking-1 is Microsoft's first fully in-house reasoning model, announced at Microsoft Build 2026 on June 2, 2026. It is part of Microsoft's MAI (Microsoft AI) model family, alongside MAI Image 2 (image generation) and MAI-Code-1-Flash (coding). MAI-Thinking-1 uses a sparse Mixture-of-Experts architecture with 35 billion active parameters and approximately 1 trillion total parameters, and supports a 256,000-token context window. Critically, it was trained from scratch on enterprise-grade, commercially licensed data — with zero distillation from OpenAI models. Microsoft reports that independent raters prefer it over Claude Sonnet 4.6, and that it matches Claude Opus 4.6 on SWE-Bench Pro coding benchmarks.
MAI-Thinking-1 is a reasoning-specialized model — it is designed to 'think' through complex problems step by step before generating a final answer, similar to OpenAI's o-series or Google's thinking models. GPT-5.5 (OpenAI's current default ChatGPT model) is a general-purpose frontier model optimized for broad instruction-following and speed. Gemini 4 is Google's most capable multimodal model, excelling at text, image, video, and audio tasks in one pass. MAI-Thinking-1 is specifically trained for high-stakes multi-step tasks: legal analysis, financial modeling, code architecture, scientific reasoning, and long-context document synthesis — where the chain of reasoning matters as much as the conclusion.
MAI-Thinking-1 is a reasoning model — it performs best when your prompt asks it to reason in steps, not just produce an answer. Effective prompt patterns include: (1) Assign a role ('You are a senior contract attorney') to set domain expertise; (2) Provide structured sub-questions ('First analyze X, then Y, then Z') to guide the reasoning chain; (3) Ask it to show reasoning before conclusions ('Walk me through your reasoning before giving a recommendation'); (4) Use explicit numbered steps in multi-part tasks; (5) For code tasks, ask it to explain WHY a bug exists, not just what to fix; (6) For analysis tasks, ask it to identify assumptions explicitly. The longer and more complex the task, the more MAI-Thinking-1 outperforms general-purpose models.
As of June 2026, MAI-Thinking-1 is available in private preview through Microsoft Foundry (Azure AI Foundry). It is also accessible via Fireworks AI, Baseten, and OpenRouter for developers who have been granted early access. Microsoft has announced a gradual general availability rollout through Azure for enterprise customers. MAI-Thinking-1 supports the standard Chat Completions API, making it a drop-in replacement for other reasoning models in existing pipelines. Consumer access via Copilot is expected to follow after the enterprise rollout.
MAI-Thinking-1 excels at tasks requiring multi-step structured reasoning: complex legal and contract analysis, financial modeling and valuation, scientific hypothesis generation and testing, software architecture review and debugging, policy impact assessment, long-context document synthesis, mathematical proofs, and multi-criteria business decisions. Its 256K-token context window means it can process entire contracts, codebases, or research corpora in a single prompt. It is specifically optimized for enterprise workflows where accuracy and traceable reasoning matter — not for creative writing, image generation, or casual conversation (where a general-purpose model is better suited).
OpenAI's default ChatGPT model — creative writing, business, coding, research
Google's most capable multimodal AI — text, image, video, audio in one pass
Google's fastest frontier model — free in Gemini app, agentic tasks, 4× speed
Microsoft's #3-ranked AI image model — available via Bing and Copilot
OpenAI's image model — instruction-following, UI design, posters, portraits
Google's most photorealistic image model — portraits, product, architecture