Jemiboo Magazine

Microsoft Ignite 2026 (November): 7 models to cut agent costs

Microsoft unveiled seven in-house AI models for coding, security, images, voice, transcription and decision-making at Ignite 2026 on November 17, 2026.

Jemiboo Magazine

· 5 min read

Read in other languages한국어Español简体中文繁體中文日本語

Key takeaways (as of November 19, 2026)

  • The coding model MAI Code 2 costs $0.50 per million input tokens and $2.50 per million output tokens until May 17, 2027, after which prices rise to $1 and $5.
  • Microsoft says more than 50 million agents are registered in its agent management tool Agent 365, which has now gained spending limits and alerts.
A navy server with seven teal status lights standing on a dark surface

More than 50 million AI agents are already registered in Agent 365, Microsoft's tool for managing agents. The figure came from the stage on the first day of Ignite 2026, held on November 17 at the Moscone Center in San Francisco. On the same day, Microsoft AI released seven in-house models at once. The event runs through November 20. The headline act at this year's Ignite was not a smarter model but a cheaper one. As agents multiply and their running costs climb, the competition has shifted to who can bring those costs down.

Seven in-house models at Microsoft Ignite 2026

Ignite is Microsoft's annual event for enterprise customers and developers. This year it opened at 10 a.m. Pacific time on November 17 with a keynote by CEO Satya Nadella. Microsoft AI, the division led by Mustafa Suleyman, then announced seven models, all in-house and all carrying the MAI name. The idea is to build a model fitted to each task and compete with OpenAI and Anthropic models on price.

ModelTaskConfirmed availability
MAI Code 2CodingGenerally available in GitHub Copilot
MAI Code 2 FlashLightweight codingGitHub Copilot; limited preview in Foundry
MAI Cyber 2 FlashSecurity analysis—
MAI Image 2.7Image generation and editingPreview
MAI Realtime 1Real-time voice conversationFrom $0.01 per minute of audio
Dragon Copilot TranscribeTranscribing clinical conversationsBuilt into Dragon Copilot
Decision 1.1Automated decision-makingGenerally available, more than 4,000 customers

The list follows Microsoft AI's November 17 announcement and model cards. Foundry is Azure's development platform where companies choose from a range of AI models. None of the seven is a general-purpose reasoning model. Instead, the lineup splits off expensive workloads such as coding and security and assigns each to its own model.

$0.50 per million tokens, with a discount caveat

The most striking detail was the price of the coding model. Until May 17, 2027, MAI Code 2 charges $0.50 per million input tokens and $2.50 per million output tokens. Tokens are the units in which AI reads and writes text. Once the discount period ends, the price doubles. Microsoft says MAI Code 2 delivers about 98% of the coding quality of GPT-6.1 Sol at about 40% of the price. According to the company's own footnote, however, that comparison is based on the 50% discounted price, and the performance figures come from benchmarks the company selected.

Model ($ per million tokens)Discounted inputDiscounted outputInput after May 17, 2027Output after May 17, 2027
MAI Code 20.502.501.005.00
MAI Code 2 Flash0.050.250.100.50

Prices are taken from the MAI Code 2 model card published on November 17, 2026. According to the same model card, MAI Code 2 has 2.8 trillion total parameters and can read 1.05 million tokens at once, enough to load an entire code repository and have it make changes. For anyone planning to use it for more than a year, it makes more sense to run the numbers on the list price after May 2027 rather than the discounted rate.

A laptop glowing teal and three coins on a dark desk

Who manages 50 million agents?

Cheaper models solve only half the problem. The other half is the growing number of agents. Agent 365 is a tool that gives each agent inside a company an identity and permissions. Microsoft has now extended its scope to agents built on Anthropic, OpenAI, Databricks and Salesforce, work that is scheduled to be completed in December. It also brings the tools agents use, such as MCP servers and plugins, under a single point of control. A cost management feature released in preview lets companies set spending policies, limits and alerts. Readers should note that the 50 million figure counts registered agents, not agents actively at work.

Speaking at Ignite on November 17, Suleyman said he keeps hearing from businesses that "token spend is going through the roof," according to Yahoo Finance. As agents take on more tasks, such as the payments covered in our November 18 article "ChatGPT Wallet (Nov 2026): AI Agents Pay Without Card Numbers" (https://jemiboo.com/en/tech/2026-11-18/), that bill grows with them.

Dates developers and businesses should watch

For developers, the first point of contact is GitHub Copilot. MAI Code 2 and Code 2 Flash are generally available and can be selected right away from Copilot's model picker. Autopilot, the always-on work agent first unveiled in September, began opening to Frontier program customers on November 17. Copilot Code, which builds dashboards and business apps from plain-language requests, is in preview on Windows, the web, iOS and Android, with general availability targeted for December.

DateMilestoneWho it applies to
November 17, 2026Seven MAI models announced; phased rollout of AutopilotFrontier program customers
November 20, 2026Ignite 2026 closes—
December 2026 (target)General availability of Copilot CodeCurrently Frontier program customers
May 17, 2027MAI Code 2 discount endsMAI Code 2 and Code 2 Flash users

Autopilot is still rolling out gradually, starting with Frontier customers. Companies using Microsoft 365 would do well to decide who controls spending limits and management permissions before adding more agents. If this event showed anything, it is the order of things: in a year when model prices fall, usage rises first.

Frequently asked questions

When and where is Microsoft Ignite 2026?

Ignite 2026 runs from November 17 to 20, 2026, at the Moscone Center in San Francisco. Keynotes and sessions can also be watched online.

How can I use MAI Code 2 and what does it cost?

MAI Code 2 is generally available in GitHub Copilot's model picker. The price on its model card is a discounted $0.50 per million input tokens and $2.50 per million output tokens until May 17, 2027, rising to $1 and $5 after that.

What is Microsoft Agent 365?

Agent 365 is Microsoft's tool for managing AI agents inside a company by giving them identities and permissions. At Ignite 2026 it gained management of other companies' agents, along with spending limits and alerts.

Sources

  1. Microsoft AI, Seven new world class models at scalable price-performance (Nov 17, 2026) — microsoft.ai
  2. Microsoft AI, MAI Code 2 and MAI Code 2 Flash Model Card (Nov 17, 2026) — microsoft.ai
  3. Microsoft, Ignite 2026: Building the frontier ecosystem (Nov 17, 2026) — blogs.microsoft.com
  4. Microsoft, Copilot at Microsoft Ignite: Expanding what's possible with intelligence built for work (Nov 17, 2026) — microsoft.com
  5. InfoWorld, Microsoft launches MAI Code 2 models to challenge OpenAI and Anthropic in AI coding (Nov 17, 2026) — infoworld.com
  6. Yahoo Finance, Microsoft's AI pitch at Ignite: Cheaper models as business 'token spend is going through the roof' (Nov 17, 2026) — finance.yahoo.com

AI agentsenterprise AIMicrosoft Ignite 2026Microsoft IgniteMicrosoft AIMAI Code 2

More from the same day

More in Tech