Other expected Build 2026 announcements provide deeper technical context for this move, including the rollout of Copilot Studio v3, the formal launch of Windows agents as an OS-level feature, and tighter Azure AI Foundry integrations for multi-model routing across systems like GPT, Claude, and open-weight models .

An important caveat: As of late May 2026, Microsoft has not yet formally revealed the coding model. Preview articles note that keynote content is unconfirmed, so granular details about the model's name, architecture, and performance metrics should be treated as expected but not guaranteed .

Microsoft's Existing AI Stack: The MAI Family

This coding model will not come out of nowhere. It is the latest and most strategically vital piece of the Microsoft AI (MAI) model family, developed in-house by the MAI Superintelligence team led by Mustafa Suleyman, the DeepMind co-founder who is now CEO of Microsoft AI .

Microsoft has already released five MAI models, starting with three foundational ones on April 2, 2026 :

• MAI-Transcribe-1: A speech-to-text engine covering 25+ languages, with Microsoft claiming it is the most accurate currently available. It operates 2.5 times faster than the company's previous standard service .
• MAI-Voice-1: An audio generation model that can produce 60 seconds of natural-sounding speech in just one second. It enables developers to create custom synthetic voices .
• MAI-Image-2: An image generation model. An updated variant, MAI-Image-2e, followed later in April 2026 .

All models are available through Microsoft Foundry, representing the company's most concrete steps to date toward owning its own AI stack rather than indefinitely renting it from OpenAI . Most importantly, the MAI team has publicly confirmed its plans to build a frontier large language model as the next step toward complete AI independence . The dedicated coding model at Build 2026 is the logical bridge to that ambition.

The Competitive Storm: Why Microsoft Had to Act

Microsoft's urgency has a clear cause: Copilot is losing its iron grip on the developer market.

Eroding Market Share

Data from the 2026 JetBrains Developer Ecosystem Survey, which polled over 10,000 developers, shows Copilot's usage share slipping from 33% to 26% in a single year . Cursor, the AI-native IDE, grabbed 18% adoption, while Anthropic's terminal-based Claude Code made a dramatic entrance at 10% in its first survey appearance . The competition has also gotten more aggressive: Cursor launched a ground-up rebuild, Cursor 3, with multi-agent orchestration in April 2026, and even integrated directly into Microsoft Teams, putting its agent on Microsoft's own collaboration turf .

The Pricing Pivot to Usage-Based Billing

The financial model for Copilot is changing overnight. Effective June 1, 2026 — the day before Build begins — GitHub Copilot transitions from flat-rate subscription billing to a metered, usage-based billing system centered on a new token-tracking currency called GitHub AI Credits .

Under the new system, subscription prices at the plan level remain the same (Pro at $10/month, Business at $19/user/month, and Enterprise/Pro+ at $39/user/month). However, those fees now buy a fixed monthly pool of AI credits. Any usage beyond that pool is an additional overage cost, calculated based on the input, output, and cached tokens consumed for each AI interaction across different models . As a result, a heavy Copilot user's bill could fluctuate significantly based on their actual compute consumption .

A Single Strategy for Three Pressures

The planned coding model is Microsoft's three-in-one response to these challenges:

1. Cost Control: Usage-based billing ties every Copilot transaction to a token price. Running a proprietary, first-party model is more cost-efficient at scale than paying per-token rates to a third-party API like OpenAI's, letting Microsoft protect its margins .
2. Performance Parity: By controlling the full model stack, Microsoft can fine-tune it precisely for the long-context, agentic, and multi-file editing workflows that have made Cursor and Claude Code so popular with developers .
3. Strategic Independence: The model reduces the company's profound technical and financial dependency on OpenAI, a relationship that has become increasingly complex as both companies race toward what they define as frontier AI .

Build 2026 is not just another product launch. It is the moment Microsoft shows a developer community increasingly tempted by other tools whether it can match their performance with a model and platform it fully controls — and charge for it in a way that makes economic sense.

Other expected Build 2026 announcements provide deeper technical context for this move, including the rollout of Copilot Studio v3, the formal launch of Windows agents as an OS-level feature, and tighter Azure AI Foundry integrations for multi-model routing across systems like GPT, Claude, and open-weight models .

An important caveat: As of late May 2026, Microsoft has not yet formally revealed the coding model. Preview articles note that keynote content is unconfirmed, so granular details about the model's name, architecture, and performance metrics should be treated as expected but not guaranteed .

Microsoft's Existing AI Stack: The MAI Family

This coding model will not come out of nowhere. It is the latest and most strategically vital piece of the Microsoft AI (MAI) model family, developed in-house by the MAI Superintelligence team led by Mustafa Suleyman, the DeepMind co-founder who is now CEO of Microsoft AI .

Microsoft has already released five MAI models, starting with three foundational ones on April 2, 2026 :

• MAI-Transcribe-1: A speech-to-text engine covering 25+ languages, with Microsoft claiming it is the most accurate currently available. It operates 2.5 times faster than the company's previous standard service .
• MAI-Voice-1: An audio generation model that can produce 60 seconds of natural-sounding speech in just one second. It enables developers to create custom synthetic voices .
• MAI-Image-2: An image generation model. An updated variant, MAI-Image-2e, followed later in April 2026 .

All models are available through Microsoft Foundry, representing the company's most concrete steps to date toward owning its own AI stack rather than indefinitely renting it from OpenAI . Most importantly, the MAI team has publicly confirmed its plans to build a frontier large language model as the next step toward complete AI independence . The dedicated coding model at Build 2026 is the logical bridge to that ambition.

The Competitive Storm: Why Microsoft Had to Act

Microsoft's urgency has a clear cause: Copilot is losing its iron grip on the developer market.

Eroding Market Share

Data from the 2026 JetBrains Developer Ecosystem Survey, which polled over 10,000 developers, shows Copilot's usage share slipping from 33% to 26% in a single year . Cursor, the AI-native IDE, grabbed 18% adoption, while Anthropic's terminal-based Claude Code made a dramatic entrance at 10% in its first survey appearance . The competition has also gotten more aggressive: Cursor launched a ground-up rebuild, Cursor 3, with multi-agent orchestration in April 2026, and even integrated directly into Microsoft Teams, putting its agent on Microsoft's own collaboration turf .

The Pricing Pivot to Usage-Based Billing

The financial model for Copilot is changing overnight. Effective June 1, 2026 — the day before Build begins — GitHub Copilot transitions from flat-rate subscription billing to a metered, usage-based billing system centered on a new token-tracking currency called GitHub AI Credits .

Under the new system, subscription prices at the plan level remain the same (Pro at $10/month, Business at $19/user/month, and Enterprise/Pro+ at $39/user/month). However, those fees now buy a fixed monthly pool of AI credits. Any usage beyond that pool is an additional overage cost, calculated based on the input, output, and cached tokens consumed for each AI interaction across different models . As a result, a heavy Copilot user's bill could fluctuate significantly based on their actual compute consumption .

A Single Strategy for Three Pressures

The planned coding model is Microsoft's three-in-one response to these challenges:

1. Cost Control: Usage-based billing ties every Copilot transaction to a token price. Running a proprietary, first-party model is more cost-efficient at scale than paying per-token rates to a third-party API like OpenAI's, letting Microsoft protect its margins .
2. Performance Parity: By controlling the full model stack, Microsoft can fine-tune it precisely for the long-context, agentic, and multi-file editing workflows that have made Cursor and Claude Code so popular with developers .
3. Strategic Independence: The model reduces the company's profound technical and financial dependency on OpenAI, a relationship that has become increasingly complex as both companies race toward what they define as frontier AI .

Build 2026 is not just another product launch. It is the moment Microsoft shows a developer community increasingly tempted by other tools whether it can match their performance with a model and platform it fully controls — and charge for it in a way that makes economic sense.