• Home
  • Technology
  • Gaming
  • Entertainment
  • World & Business
  • Science
  • Sports
  • AI
HomeTechnologyGamingEntertainmentWorld & BusinessScienceSportsAI
AI
Report

Microsoft Launches Decision-1, Low-Cost Specialist Model for Agent Control and Routing

Microsoft released Microsoft-Decision-1 (post-trained on Qwen3.5-9B), a classifier model outputting probabilities for predefined options.

4 Sources, 2h ago, first seen 2h ago

TLDR

Microsoft-Decision-1 is priced at $0.042 per million input tokens, with reported accuracy of about 83.5% and latency of 85ms. A post argues that comparing it with Jev and Clef through a shared LiteLLM entry point requires testing fixed inputs, recording model versions, latency and cost, checking edge cases and setting rollback conditions—not just adding a routing option.

Combined views

—

4 Sources, first seen 2h ago

— likes— comments— saves— reposts

Combined views

—

4 Sources, first seen 2h ago

— likes— comments— saves— reposts

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

Sentiment

Positive——Negative

Summary

Not enough discussion yet.

No sentiment analysis available yet.

4 Sources

星星之火可以燎原@fireandstart把 Jev、Clef、Microsoft-Decision-1 放到同一个 LiteLLM 入口里,真正有用的地方不只是少配几个 endpoint,而是能把“选哪个模型处理这类决策”变成可比较、可复查的工程问题。Decisions playground 提供了一个开始做对照的地方,但接入本身并不等于路由已经选对:同一批固定输入要看结果是否符合任务目标,也要记录模型版本、延迟和成本,特别关注少数边界案例是否被平均分掩盖。再把这些结果连到实际调用链,才知道切换模型带来的是稳定提升,还是只在演示样例上好看。模型越多,越需要把选择依据、评估集和回退条件留在团队能复核的地方。特别是决策模型的输出往往影响后续动作,不能只看文本是否说得通,还要定义什么叫正确、如何处理不确定或无结论的输出,以及线上表现变差时怎样快速回滚。否则界面多了一项路由功能,团队却仍无法解释当初为什么选它。对开发者来说,这类统一入口若能缩短试验路径,同时保留清楚的测量与决策记录,就比单纯“能调用更多模型”更有价值。2h
VB@techtacticianSatya Nadella says frontier AI should be treated like an insider risk: assume it can be compromised, log its actions, and give humans an "emergency brake" to pause or stop a model mid-task. Trust has to live outside the model. #AI Source: https://www.cnbc.com/2026/10/10/microsoft-satya-nadella-ai-emergency-brake-safety.html1h
ArtoVista@artovistaMicrosoft-Decision-1 doesn't write. Give it a question and a fixed list of options. It picks one, with a confidence signal. Xbox Research: quality competitive with GPT-5, 80-100x faster. Built on Qwen3.5-9B. Microsoft's own numbers, preview only. Source: Microsoft Foundry Blog1h
Ranferis Morales@Ranferis_MMicrosoft is turning the labs into suppliers: - Copilot Researcher: GPT writes, Claude checks - Decision-1: built on Qwen - Satya: "separate the supply of intelligence from the authority over it" Labs are racing for the best model. Microsoft is making sure it doesn't matter1h
    • Home
    • Technology
    • Gaming
    • Entertainment
    • World & Business
    • Science
    • Sports
    • AI

    4 Sources

    星星之火可以燎原@fireandstart把 Jev、Clef、Microsoft-Decision-1 放到同一个 LiteLLM 入口里,真正有用的地方不只是少配几个 endpoint,而是能把“选哪个模型处理这类决策”变成可比较、可复查的工程问题。Decisions playground 提供了一个开始做对照的地方,但接入本身并不等于路由已经选对:同一批固定输入要看结果是否符合任务目标,也要记录模型版本、延迟和成本,特别关注少数边界案例是否被平均分掩盖。再把这些结果连到实际调用链,才知道切换模型带来的是稳定提升,还是只在演示样例上好看。模型越多,越需要把选择依据、评估集和回退条件留在团队能复核的地方。特别是决策模型的输出往往影响后续动作,不能只看文本是否说得通,还要定义什么叫正确、如何处理不确定或无结论的输出,以及线上表现变差时怎样快速回滚。否则界面多了一项路由功能,团队却仍无法解释当初为什么选它。对开发者来说,这类统一入口若能缩短试验路径,同时保留清楚的测量与决策记录,就比单纯“能调用更多模型”更有价值。2h
    VB@techtacticianSatya Nadella says frontier AI should be treated like an insider risk: assume it can be compromised, log its actions, and give humans an "emergency brake" to pause or stop a model mid-task. Trust has to live outside the model. #AI Source: https://www.cnbc.com/2026/10/10/microsoft-satya-nadella-ai-emergency-brake-safety.html1h
    ArtoVista@artovistaMicrosoft-Decision-1 doesn't write. Give it a question and a fixed list of options. It picks one, with a confidence signal. Xbox Research: quality competitive with GPT-5, 80-100x faster. Built on Qwen3.5-9B. Microsoft's own numbers, preview only. Source: Microsoft Foundry Blog1h
    Ranferis Morales@Ranferis_MMicrosoft is turning the labs into suppliers: - Copilot Researcher: GPT writes, Claude checks - Decision-1: built on Qwen - Satya: "separate the supply of intelligence from the authority over it" Labs are racing for the best model. Microsoft is making sure it doesn't matter1h
    Today's Rank

    —

    Not ranked yet

    Today's Rank

    —

    Not ranked yet