Microsoft MAI Image 2.7 and Flash enter public preview with Canvas | Magica
Microsoft expands MAI Image 2.7 access, adds cheaper Flash and Canvas
Editorial Team
••📖7 min read
Microsoft expands MAI Image 2.7 from an editing preview into image generation and API access, with cheaper Flash pricing and a Canvas workspace. Benchmark claims depend on the workload, while the MAI Labs interface carries restrictions on commercial use.
MAI Image 2.7 and Flash add public-preview API access and text-to-image generation to an earlier editing preview.
Microsoft advertises roughly $35 per 1,000 images for the full model and $10 for Flash, with resolution and input charges affecting the bill.
Canvas runs in MAI Labs, whose personal, non-commercial-use restriction matters for people considering it for work.
Microsoft AI, Microsoft's in-house model operation, expanded its MAI Image 2.7 offering on November 17 with text-to-image generation, public-preview API access and a lower-cost Flash variant. It also introduced Canvas, an interactive image workspace powered by Flash, in its launch announcement.
The expansion gives developers another internally built Microsoft option alongside models from outside suppliers such as OpenAI, on which Microsoft has long relied for advanced capabilities. But the broader strategy remains task specialization: reporting on the Ignite releases notes that Microsoft's comparisons use selected benchmarks and that some pricing advantages across the releases reflect promotional discounts. The image launch does not establish parity across general-purpose AI tasks.
MAI Image 2.7 first appeared on October 27 as an image-editing preview. In that introduction, Microsoft said it improved on 2.6 across all 10 use cases and seven editing actions evaluated by Artificial Analysis, which evaluates image-model preferences. Its strongest category was scene and style, where Microsoft reported a 24-Elo lead over the next-best model. The preview announcement promised text-to-image generation later; that capability and API distribution are consequential additions now.
The two-tier approach also predates this launch. The Microsoft model catalogue dates both MAI Image 2.6 and 2.6 Flash to September 4. Those predecessors already offered generation, editing and five-image references. The new family improves Microsoft's existing product ladder rather than introducing the choice between precision and speed.
Microsoft is pitching Copilot Workspaces as a shared environment for people and AI agents. Team collaboration is already in early access, but personal Autopilot integration is still coming soon, with final agent pricing unannounced.
Editorial Team
Both current variants generate images from text and edit existing images. They support up to five reference images, allowing combinations of people, products, styles and scenes, according to the model listing. The full model is the precision tier; Flash prioritizes lower latency and cost.
The maximum output is now 2048 × 2048 pixels, and both models support transparent backgrounds, according to the gateway release details. Those features matter for product images, logos and other assets that need to fit into a larger layout. Microsoft says targeted edits can change an object's color, material or position while preserving the surrounding image.
Canvas packages that editing capability into a workspace in MAI Labs. Users can begin with a prompt or sketch and select parts of an image to replace, move or restyle. The distinction is practical: the API supplies image generation and editing to developers, while Canvas gives users an interface for successive changes in one place.
Access conditions differ. Although Microsoft's model page promotes commercial workflows, its MAI Labs entry notice limits that experimental service to personal, non-commercial use and expressly excludes school or work activities. Canvas's availability in Labs should therefore not be treated as permission to use it for client work. That restriction describes Labs; the separate fal API listing explicitly labels its offering for commercial use.
The opening screen of Microsoft AI's Canvas demonstration. Canvas runs in MAI Labs, which restricts use to personal, non-commercial activities. Source: Microsoft AI.
Lower prices, with a workload attached
Microsoft's roughly $35 and $10 headline prices are per 1,000 images. Its accompanying comparison specifies 1024 × 1024 pixels; they should not be read as universal prices for every resolution or editing request.
The underlying billing is token-based. The Flash API details list $1.75 per million text input tokens, $2.50 per million image input tokens and $10 per million image output tokens. A 1024 × 1024 output uses 1,024 image tokens and costs $0.01024 before input charges, with the total request rounded up to the nearest $0.0001. OpenRouter lists the full model at $5 per million text input tokens, $8 per million image input tokens and $34 per million image output tokens. Reference-heavy editing therefore has costs beyond the headline output price.
Microsoft's launch chart supplies a useful comparison within the same editing evaluation:
Model
Editing quality, Elo
Price per 1,000 images at 1024 × 1024
MAI Image 2.7
1,215
$34.80
MAI Image 2.7 Flash
1,179
$10.24
MAI Image 2.6
1,150
$38.91
MAI Image 2.6 Flash
1,145
$19.46
GPT Image 2.6 Sunburst
1,211
$210.72
Step Image Edit 2
1,008
$3.00
Source: Microsoft's published comparison, using Artificial Analysis AA-Image-Editing v2.1 competitor figures retrieved November 16 and released production figures for the new MAI models.
On those inputs, the new Flash costs about 47% less than 2.6 Flash while scoring higher. But MAI is not the cheapest option in the chart: Step Image Edit 2 costs less and scores substantially lower. The choice is a quality-cost trade-off, rather than a single winner on price.
There is also a discrepancy within Microsoft's presentation. Its headline says the full model costs one-fifth as much as GPT Image 2.6 Sunburst, while the chart's $34.80 versus $210.72 works out to about one-sixth. Neither ratio should be generalized beyond the specified configurations and workload; Microsoft itself cautions that performance, speed and cost vary by methodology.
Company-reported prices for selected models in Microsoft's editing comparison at 1024 × 1024 pixels. Price alone does not establish equivalent editing quality. Source: Microsoft AI.
A narrow editing lead is not a universal quality verdict
Microsoft says in its evaluation notes that MAI Image 2.7 ranks first on Artificial Analysis's editing evaluation as of November 16, with 1,215 Elo across 5,478 votes. Elo summarizes relative preference in the evaluation. The reported lead over OpenAI's GPT Image 2.6 Sunburst (max) is four Elo; Microsoft supplies no uncertainty interval in these notes. That result does not establish a reliable advantage for every editing task.
The distinction between a model and its developer also matters. Microsoft's Ignite announcement says 2.7 ranks fourth in text-to-image generation while making Microsoft the second-ranked lab. Describing Microsoft as a top-two lab should not be mistaken for a second-place model ranking.
An early four-prompt test offers a smaller, practical check on the two tiers. At 1,152 × 768 pixels, the full model took 29–38 seconds per image and cost $0.0296, against 14–17 seconds and $0.0087 for Flash. Both handled the tested typography and product imagery, while the full model preserved more detail and more coherent small signs in a crowded night-market scene.
Flash being faster than its sibling does not establish an improvement over the previous generation. The same report's early OpenRouter averages put 2.7 at about 33 seconds and Flash at 18 seconds, versus about 24 and 10 seconds for their predecessors. Those small, first-day samples are not a controlled comparison. The report also predates Microsoft's launch announcement, so its statements that official launch details and generation rankings were still missing have been overtaken by Microsoft's disclosures.
What users still need to establish
Distribution gives Microsoft a route beyond benchmark attention. Its launch notes say MAI Image became the default model family in Bing Image Creator after June's Build event, with usage growing at a reported compound weekly rate of 14.7%. The announcement supplies no absolute usage baseline, so that family-wide growth rate does not establish market share or the new models' adoption. Microsoft also names Figma Weave, Minute, Wirestock and Manus as working with MAI Image; that description does not quantify paid deployments.
The models are available through Microsoft's Foundry platform and API partners including fal, Genspark, OpenRouter and Vercel. Vercel says its gateway lets developers generate and edit with one API key without a Microsoft account. Yet an intermediary does not necessarily diversify the underlying infrastructure: the OpenRouter listing shows Microsoft's Azure cloud as the sole host for the full model, forwarding requests without provider routing. It also says EU in-region routing is unavailable, a deployment constraint for customers who need that route.
For designers and application builders, the next decision is which permitted deployment route and model tier can deliver an acceptable finished asset. Testing needs to count reference-image inputs, resolutions and repeated edits, alongside latency and regional requirements. Flash's lower output price is established; whether it lowers the cost of finished work depends on how many attempts and revisions each job requires.
Google is rolling out file references and sharing for Gemini skills as adult personal accounts lose the ability to use Gems. Automatic conversion preserves supported material, but file limits and missing tools can interrupt existing workflows.
Muddy Waters disclosed a short position in Nvidia while questioning a possible connection involving senior adviser Ajay K. Puri. Its own account leaves the key identity question unanswered, separating its new claims from an existing server-diversion prosecution.
Nvidia reported $115.8 billion in quarterly revenue in its first earnings release since co-founder Jensen Huang resigned. Strong demand gives interim CEO Colette Kress a financial cushion, while a campaign for Huang’s return collides with unresolved investigations and the board’s reported opposition.
OpenAI launched GPT-6.2 Astra and Sol for professional work, cutting Astra’s standard input and output token prices by 20%. The models offer cheaper paths through selected tasks, but research evaluations, reporting failures and unfinished independent safeguard testing complicate the case for broader autonomy.
Microsoft’s first live speech-to-speech model offers OpenAI API compatibility and planned enterprise integrations. Its published evaluations show a low-cost option, with several rivals completing more agentic tasks.
Canberra-based Trellis Data is offering its 220-billion-parameter Agentaus 1 model to business and government. Its pitch goes beyond keeping data in Australia, but performance, deployment costs and a planned capital raising remain key tests.
Perplexity is adding direct collaboration between frontier models to Computer, letting a lead agent coordinate up to 16 supporting agents. Company-reported benchmarks suggest mixed teams can outperform repeated copies of one model, but leave customer costs and production reliability unresolved.
Marsh reported 105 client cyber incidents in the last two weeks of October, against 48 in the final two weeks of May. The broker links the increase partly to AI, while warning that insurance panel access may not guarantee an available response team.
Meta’s Muse personal AI agent has launched in Australia, with some local shopping and booking connections to follow. Its browser and payment tools broaden its reach, while the company’s technical disclosures qualify its privacy promises.
Microsoft and Salesforce are extending their sales software alliance with Copilotforce, combining CRM workflows with Microsoft 365 context under existing permissions. The Copilot plugin enters public preview, while Teams remains in pilot; earlier Microsoft tools and a Claude beta raise the bar for proving its value.
OpenAI is widening applications for its legal AI configuration after improving research scores and expanding its search index. Access still requires approval, and benchmark results show substantial limits alongside the gains.
Microsoft has made Word, Excel and PowerPoint creation generally available inside Copilot, combining editable files with workplace permissions and collaboration. The broader shift toward autonomous work remains staggered across previews, while its claimed cost advantage rests on company-funded testing.
Knit Health’s NEA-led Series A funds deployment of Med-BEAM for care navigation and hospital operations. Its early company benchmark beats three Claude models at predicting clinical actions, while better patient outcomes remain unproven by that test.
Science Corporation’s Series D follows its European retinal-implant launch. Clinical results support improved central vision, but U.S. approval, eligible patients and commercial terms remain consequential questions for the startup, reportedly valued at $3.5 billion.
Google will let Gemini users save conversations while opting out of generative AI training on future chats. The global rollout also extends storage of history-off and temporary chats from 72 hours to around two months; human review and longer retention can still apply.
Google is removing the subscription requirement for Gemini’s task mode in six countries, while adding U.S. payment, calling and messaging features. The expansion broadens access to delegated work, but usage limits, permissions and website compatibility still govern what it can finish.
Eligible Pixel owners can join a Play Store beta to generate home-screen widgets from descriptions, but device, language and regional restrictions still apply.
Microsoft’s new in-house coding models compete with outside suppliers inside GitHub Copilot. Company tests show lower task costs at promotional prices, while its continuing financial ties to OpenAI complicate the rivalry.
Valkryn disclosed a $15 million seed round and early military testing contracts. Its evaluator-focused Battle Lab faces existing commercial tools, public testing capabilities and funding constraints that software alone cannot resolve.