Microsoft
Washington, संयुक्त राज्यMAI Image 2.5 Flash is the ultra-efficient variant of Microsoft's MAI Image 2.5 family for lower-cost text-to-image and image editing workflows. This entry tracks the Build 2026 announcement.
MAI Thinking 1 is Microsoft's flagship reasoning model for chat, code generation, and complex multi-step workflows. This entry tracks the Build 2026 launch ahead of broader public availability.
MAI Voice 2 Flash is the lower-cost, ultra-efficient variant of Microsoft's MAI Voice 2 text-to-speech family. This entry tracks the Build 2026 announcement ahead of release.
Decision-scoring model for routing, classification, prioritization, verification, and workflow control. Scores fixed answer options with calibrated probabilities.
MAI-Transcribe-2-Streaming is a low-latency speech-to-text model from Microsoft for real-time transcription. Audio is sent as a continuous stream and incremental transcripts are returned while the speaker talks, with intermediate results updated and final results confirming each segment.
MAI-Voice-2.1 is a text-to-speech model from Microsoft for expressive, high-fidelity speech generation. It supports 23 languages, SSML emotion and style control, gated instant voice cloning, and long-form generation that keeps speaker consistency across audiobooks, podcasts and narration.
MAI-Voice-2.1-Flash is a text-to-speech model from Microsoft for low-latency, real-time speech generation. It produces expressive speech across 23 languages with emotion and style control through SSML, and supports gated instant voice cloning.
MAI-Image-2.6 is a diffusion-based image generation and editing model from Microsoft for text-to-image generation and image-to-image edits. Prompts can be up to 32,000 tokens, edits accept up to five reference images, and output is PNG up to 2,359,296 total pixels.
MAI-Image-2.6-Flash is a diffusion-based image generation and editing model from Microsoft with the same text-to-image and image-to-image capabilities as MAI-Image-2.6, offered as a faster variant. Prompts can be up to 32,000 tokens, and edits accept up to five reference images.
MAI-Transcribe-2 is a speech-to-text model from Microsoft for transcribing audio files across 60 languages. It provides speaker diarization, word-level timestamps, automatic language identification, keyword biasing, code switching, and clean or verbatim transcription styles.
MAI Code 1.1 Flash is Microsoft's efficient agentic coding model for GitHub Copilot, with improved CLI and .NET performance over MAI Code 1 Flash.
MAI Cyber 1 Flash is Microsoft's compact specialist cybersecurity model for finding vulnerabilities in complex codebases. Access is restricted to verified defenders through MDASH.
MAI Image 2.5 is Microsoft's image generation model for both text-to-image and image editing workflows. Microsoft announced it at Build 2026 and described the image family as generally available in Microsoft Foundry and MAI Playground.
MAI Transcribe 1.5 is Microsoft's speech-to-text model for production transcription workflows. Microsoft announced broad language support, domain-specific terminology handling, and general availability in Microsoft Foundry and MAI Playground at Build 2026.
MAI Voice 2 is Microsoft's multilingual text-to-speech model for natural spoken output with short-sample voice adaptation. Microsoft positioned the voice family as generally available in Microsoft Foundry and MAI Playground at Build 2026.
Phi Ground Any is Microsoft's broader GUI grounding model for click-point localization and computer-use tasks across arbitrary screen layouts.
MAI Image 2e is Microsoft's more efficient text-to-image variant of MAI Image 2, built for lower-cost production generation.
MAI Transcribe 1 is Microsoft's multilingual speech-to-text model for production transcription across 25 major languages.
MAI Voice 1 is Microsoft's text-to-speech model for fast, natural spoken output in production voice experiences.
MAI Image 2 is Microsoft's text-to-image model for high-quality creative generation with strong photorealism and in-image text rendering.
Phi 4 Reasoning Vision 15B is Microsoft's multimodal open-weight reasoning model for visual question answering, OCR, chart reasoning, and computer-use grounding tasks.
Phi Ground is Microsoft's GUI grounding model, finetuned from Phi 3.5 Vision Instruct for element localization and computer-use agent workflows.
Phi Mini MoE Instruct is Microsoft's open-weight mixture-of-experts instruction model with 7.6B total parameters and 2.4B active parameters.
Phi Tiny MoE Instruct is Microsoft's smaller open-weight mixture-of-experts instruction model with 3.8B total parameters and 1.1B active parameters.
Phi 4 Mini Reasoning is a reasoning model from Microsoft for multi-step analysis, math, and tool-driven tasks. It is a smaller-footprint option suited to tighter latency or deployment budgets.
Phi 4 Reasoning is a reasoning model from Microsoft for multi-step analysis, math, and tool-driven tasks.
Phi 4 Reasoning Plus is a reasoning model from Microsoft for multi-step analysis, math, and tool-driven tasks.
Phi 4 Mini is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data.
Phi 4 multimodal instruct is a multimodal model from Microsoft that can work across text and visual inputs.
Phi 4 is designed to perform well in complex reasoning tasks and can operate efficiently in situations with limited memory or where quick responses are needed.
Phi 3.5 mini instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data.
Phi 3.5 MoE instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data.
Phi 3.5 vision instruct is a multimodal model from Microsoft that can work across text and visual inputs.
Phi 3 Mini 4K Instruct is Microsoft's compact 3.8B open-weight instruction model with a short 4K context window optimized for efficient reasoning and coding workloads.
WizardLM-2 8x22B is Microsoft AI's most advanced Wizard model. It demonstrates highly competitive performance compared to leading proprietary models, and it consistently outperforms all existing state-of-the-art opensource models.
50 layers deep image classification CNN trained on more than 1M images from ImageNet
MAI Code 1 Flash is Microsoft's inference-efficient agentic coding model, tailored for GitHub Copilot and Microsoft developer workflows.
Phi 1 is designed to perform well in complex reasoning tasks and can operate efficiently in situations with limited memory or where quick responses are needed.
Phi 1.5 is designed to perform well in complex reasoning tasks and can operate efficiently in situations with limited memory or where quick responses are needed.
Phi 2 is designed to perform well in complex reasoning tasks and can operate efficiently in situations with limited memory or where quick responses are needed.
Phi 3 Medium 128K Instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data.
Phi 3 Medium 4K Instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data.
Phi 3 Mini 128K Instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data.
Phi 3 Small 128K Instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data.
Phi 3 Small 8K Instruct is a lightweight open model built upon synthetic data and filtered publicly available websites - with a focus on high-quality, reasoning dense data.
Phi 3 Vision 128K Instruct is a multimodal model from Microsoft that can work across text and visual inputs.
Phi 4 Mini Flash Reasoning is a reasoning model from Microsoft for multi-step analysis, math, and tool-driven tasks. It is tuned for lower latency and cost-sensitive production use.