Microsoft
Phi-4-multimodal-instruct is a open-weight Microsoft speech audio model.
Running this yourself: consumer gpu should be enough.
39.4
Quality Score
---
Arena ELO
6B
Parameters
---
Context
This measures the amount of verifiable public evidence we have, not how capable the model is. A missing field means it has not been verified yet, not that its value is zero.
14 of 22 public signals
Sign in to join the discussion
362.6K
Downloads
1.6K
Likes
Feb 2025
Released
3/5 signals
3/4 signals
3/5 signals
3/4 signals
2/4 signals
Gaps we are still tracking
Launches
1
Benchmarks
2
Research
2
General
6
Recent launch, pricing, benchmark, and API signals linked to this model or its provider.
Thrilled to announce by popular demand MAI-Code-1-Flash is now generally available for GitHub Copilot Business and GitHub Copilot Enterprise - fast, efficient, and custom designed to help you build more for less on @github
View source
Quality: 4.2/100 | Price: $0/M tokens | Output: 17.547 tok/s | MMLU: 0.485% | HumanEval: 0.131%
Quality: 4.2/100 | Price: $0/M tokens | Output: 17.547 tok/s | MMLU: 0.485% | HumanEval: 0.131%
View sourceRecent agent benchmarks increasingly ground evaluation in executable environments, from code repair to web navigation, app APIs, and function calling. Yet completing consequential work beyond code requires more than producing a plausible response or valid tool call: agents must gather missing information over multiple turns, follow domain policies, coordinate dependent tools, and realize the correct persistent state transition without collateral effects. In this paper, we introduce Thinkingbox, a sandbox for tool-agent-user interaction that provides isolated MCP-compatible tool sessions, complete execution traces, and outcome evaluation over terminal backend state. Built on this sandbox, Thinkingbox-bench contains 507 policy-conditioned workflows across numerous scenarios, including retail, hospitality, auto insurance, neobank internal IT, and consulting IT/HR support. Each attempt is evaluated by task-specific executable checks that accept valid trajectories while rejecting wrong, missing, or extra effects; designated tasks additionally check required properties of the final response. Across proprietary and open-weight models, the strongest achieves 65.36% pass@1, but only 25.25% pass^20. Moreover, many failed trials show clean termination and valid state-changing actions, showing that response or tool-call-level signals are not clear proxies for end-to-end task completion. Thinkingbox-bench reveals a large gap between occasionally finding a successful trajectory and reliably completing stateful business tasks. We release both Thinkingbox and Thinkingbox-Bench: https://github.com/microsoft/thinkingbox
View sourceThe Muon optimizer incurs a significant overhead cost due to its cubic-time Newton-Schulz orthogonalization step. When weights are sharded, communication overhead compounds this computational cost, eroding the benefits of Muon in many settings. We present Dion3, a revision of Muon that targets this overhead at every level of the stack. Our Gram Newton-Schulz algorithm reduces the FLOP cost of orthogonalization, our CuteDSL kernels accelerate it by exploiting symmetry, and our megabatching strategy reduces communication overhead. Moreover, we propose a simple change to the update rule that cuts costs even further: selecting only a fraction of the momentum matrix's rows to orthogonalize at each step. This update rule improves on Dion (another "compressed" version of Muon), in both speed and performance. Overall, Dion3 matches or improves on the loss achieved by Muon but reduces optimizer step time by up to 6x. Dion3 is available via the dion package (https://github.com/microsoft/dion) as a drop-in replacement for Muon.
View source
Say how you feel. Get a poem that meets you there. Ode connects you with a poem for your moment. Guided by William Sieghart. Powered by Microsoft AI models. Try it out here: https://t.co/egu88NTf9C https://t.co/jQ3qm9sKnz
Thrilled to announce by popular demand MAI-Code-1-Flash is now generally available for GitHub Copilot Business and GitHub Copilot Enterprise - fast, efficient, and custom designed to help you build more for less on @github
We shipped a new coding model built for your everyday dev work. MAI-Code-1-Flash is fast, token-efficient, and trained inside real GitHub Copilot environments. It plans, builds, runs, and tests. All from Copilot Chat in VS Code. Watch it go from a single frost banner to a full
MAI-Image-2.5 ranked #2 for text-to-image and #3 for image editing on @ArtificialAnlys - showing strong performance across both generation and precise image edits. From rainy-window blur to a clear, usable street scene, while preserving object consistency, lighting, https://t.co/q8MFDrr8uB
What happens when speech, transcription, and coding models work together? This prototype demo, built using a VS Code fork, showcases how MAI-Transcribe, MAI-Voice, and MAI-Code-1-Flash can work together in a unified workflow to transform spoken instructions into working code. https://t.co/z35l4fV2xQ
Behind every model is a team dedicated to solving difficult challenges, exploring new ideas, and continuously pushing technology forward. Meet some of the people behind Microsoft AI. Watch the full video here: https://t.co/oNxtle7FtO https://t.co/8fsVY6DnFQ
Recent agent benchmarks increasingly ground evaluation in executable environments, from code repair to web navigation, app APIs, and function calling. Yet completing consequential work beyond code requires more than producing a plausible response or valid tool call: agents must gather missing information over multiple turns, follow domain policies, coordinate dependent tools, and realize the correct persistent state transition without collateral effects. In this paper, we introduce Thinkingbox, a sandbox for tool-agent-user interaction that provides isolated MCP-compatible tool sessions, complete execution traces, and outcome evaluation over terminal backend state. Built on this sandbox, Thinkingbox-bench contains 507 policy-conditioned workflows across numerous scenarios, including retail, hospitality, auto insurance, neobank internal IT, and consulting IT/HR support. Each attempt is evaluated by task-specific executable checks that accept valid trajectories while rejecting wrong, missing, or extra effects; designated tasks additionally check required properties of the final response. Across proprietary and open-weight models, the strongest achieves 65.36% pass@1, but only 25.25% pass^20. Moreover, many failed trials show clean termination and valid state-changing actions, showing that response or tool-call-level signals are not clear proxies for end-to-end task completion. Thinkingbox-bench reveals a large gap between occasionally finding a successful trajectory and reliably completing stateful business tasks. We release both Thinkingbox and Thinkingbox-Bench: https://github.com/microsoft/thinkingbox
The Muon optimizer incurs a significant overhead cost due to its cubic-time Newton-Schulz orthogonalization step. When weights are sharded, communication overhead compounds this computational cost, eroding the benefits of Muon in many settings. We present Dion3, a revision of Muon that targets this overhead at every level of the stack. Our Gram Newton-Schulz algorithm reduces the FLOP cost of orthogonalization, our CuteDSL kernels accelerate it by exploiting symmetry, and our megabatching strategy reduces communication overhead. Moreover, we propose a simple change to the update rule that cuts costs even further: selecting only a fraction of the momentum matrix's rows to orthogonalize at each step. This update rule improves on Dion (another "compressed" version of Muon), in both speed and performance. Overall, Dion3 matches or improves on the loss achieved by Muon but reduces optimizer step time by up to 6x. Dion3 is available via the dion package (https://github.com/microsoft/dion) as a drop-in replacement for Muon.