Microsoft Adds Meta's Muse Models to Microsoft Foundry, Escalating Platform Competition
At its Ignite conference in San Francisco on November 17, 2026, Microsoft announced that three of Meta's latest Muse AI models—Muse Wonder, Muse Spark 1.3, and Muse Ember—are "coming soon" to Microsoft Foundry, the company's platform for building and running AI applications and agents [1][4][5]. The move places models from two frontier AI competitors side by side on the same cloud service, alongside OpenAI's GPT series, Anthropic's Claude, and Microsoft's own rapidly expanding MAI family.
A platform play, not just a model deal
Microsoft's Foundry catalog already lists more than 11,000 models from providers including Microsoft, OpenAI, Anthropic, Meta, Mistral, the open-source community and others [1]. The addition of Meta's newest agentic models extends a strategy Microsoft has articulated for months: give customers the flexibility to choose the right model for each task without locking into a single lab. "Their expertise should strengthen their business, not deepen their dependence on any one AI lab," the Foundry blog post states [1].
Both CEOs framed the announcement as a win for choice. Satya Nadella posted on X that Microsoft is "giving developers and enterprises even more choice and flexibility," while Mark Zuckerberg separately confirmed the three Muse models were "coming soon" to Foundry [5].
What the Muse models are positioned to do
Microsoft's blog describes the three models as a tiered set covering frontier long-horizon planning and complex coding, a balanced tier, and fast, cost-efficient execution at scale [1]. The post says the trio gives teams "frontier, balanced and high-efficiency options to match each step of an agentic workflow" [1].
No release date, pricing, or regional availability details were provided. The models are listed as "coming soon" in the Foundry model catalog [1].
Microsoft's own models launch in parallel
The Meta announcement was one piece of a much larger model rollout at Ignite. Microsoft introduced or updated its entire MAI family in a single day:
- MAI-Code-2 and MAI-Code-2-Flash for agentic coding and high-volume everyday coding, respectively [1]
- Microsoft-Decision-1.1, a Gemma-based model for routing, classification, and verification, now generally available with image input support [1]
- MAI-Image-2.7 and MAI-Image-2.7-Flash for creative media generation, in public preview [1]
- MAI-Cyber-2-Flash for vulnerability discovery and remediation, available through the Codename MDASH security harness [1]
- MAI-Thinking-1, a 35B-parameter sparse mixture-of-experts reasoning model, in preview [2]
Microsoft also announced partnerships to bring Baseten-optimized open models—Kimi-K3, GLM 5.3, GLM-5.3-Flash and DeepSeek V4.1-Flash-1017—as a Microsoft first-party service, and fal's video models—Kling v3 Pro, MiniMax H3 Max, MiniMax H3 Max Turbo, LTX-2.5 Pro and LTX-2.5 Fast—to Foundry Models through a partnership [1].
Benchmark claims are self-reported
Microsoft's performance claims for its new models come from internal benchmarks. The company says MAI-Code-2 delivers 98% of GPT-6.1 Sol's performance at 40% of the cost, and 96% of Claude Opus 5.5 at 5% of the cost, averaged across four coding benchmarks [3][4]. For MAI-Cyber-2-Flash, Microsoft cites its "Cyvergime E2E" evaluation showing less than 2% of GPT-6 Astra's cost per task at equivalent quality [4].
These figures are Microsoft's own. As the InfoWorld analysis notes, "treat vendor-run benchmarks as a starting hypothesis and test the model on your own workloads" [2]. Several MAI models remain in preview, meaning support guarantees and change expectations differ from generally available releases [2].
Hardware and infrastructure underneath
The model push is backed by Microsoft's custom silicon. The Maia 300 inference accelerator, described as a full platform spanning rack, server, and networking, is said to deliver 2.5x high-bandwidth memory performance over its predecessor [2]. Microsoft also announced the Cobalt 200 VM—its own Arm-based CPU—will expand to 23 Azure regions by year-end, with up to 50% better performance than Cobalt 100 on some AI workloads [5].
Managed Compute, a dedicated infrastructure option for running open and custom models on Foundry, reached general availability at Ignite [1].
Competitive context
The Meta partnership is notable because both companies are building frontier models. Microsoft, meanwhile, has invested heavily in OpenAI while simultaneously building its own MAI family and hosting competitor models. Foundry now positions itself as a neutral layer: Microsoft models, partner models, and open models all deployable through the same control plane, with governance tied to Entra, Defender, and Purview [1][2].
What remains unknown: when the Muse models will actually ship, what they will cost, and how they perform on independent evaluations. Microsoft's blog says only "coming soon." Until then, the announcement is a signal of intent—Microsoft wants Foundry to be the default marketplace for enterprise AI, and Meta wants distribution for its newest models—rather than a shipping product developers can use today.
Sources
- From model choice to your durable advantage: Build AI that keeps ... (techcommunity.microsoft.com)
- Microsoft MAI Models and Maia 300: The Push for Lower-Cost AI on Azure (windowsforum.com)
- Microsoft Ignite: new MAI models, Azure agent tooling... (daily.dev)
- Microsoft (MS) is competing with OpenAI and Antropic by unveiling its ... (mk.co.kr)
- Microsoft to add three Meta AI models to Foundry (breakingthenews.net)