Nace AI Open-Sources Drex 1.5: A 9B Decision Model That Scores Options, Not Text
https://www.marktechpost.com/2026/10/09/nace-ai-open-sources-drex-1-5-a-9b-decision-model-that-scores-options-not-text/📌 【Nace.AI 開源】Drex 1.5：9B 決策模型不寫文字，只對選項打分數
TL;DR：Drex 1.5 是開源的 9B 決策模型，單次 forward pass 就能在固定選項中給出機率，可直接替換商用的 Jev。
當所有人都在比拼 LLM 寫文章、寫程式碼有多流暢時，Nace.AI 選擇做一件反直覺的事：讓模型完全不生成文字，只負責在你給定的選項裡打分數。
🤔 這不是聊天模型，是決策層
Drex 1.5 讀取一個 state（文字或 JSON）加上多個具名問題，對每個問題回傳固定選項的機率分布。它支援三種問題型態：choice（多選一）、noul（是非題）與 ordinal score（順序分數）。因為不採樣 token，temperature 和 top_p 這類參數完全不適用，模型也只能回答你提供的選項，不會自由發揮。它對外服務 POST /v1/systemone API，與催生這個類別的closed model「Jev」（由 TypeSafe 開發）用的是同一套請求格式，Nace 表示現有的 Jev 客戶端只要改幾個環境變數就能切換過來。
🧩 架構：蒸餾骨幹 + 獨立指標頭
Drex 1.5 的 backbone 是 MiMo-V2.6-Distill-Qwen-9B，由 Qwen 3.5 9B 蒸餾而來，共 32 層，採用混合 attention：每 1 層全注意力搭配 3 層 linear attention。模型另外接了一個獨立的 pointer head（head.pt），直接從 backbone 的隱藏狀態為每個選項打分。每個問題只需要對「state + 該問題」跑一次 forward pass；在 llama.cpp 實作中，state 只編碼一次，可在多個問題間共用，節省重複運算。Nace 表示訓練資料來自各項 Decision Index benchmark 的官方訓練切分，評測則只用 held-out 切分。
📊 排行榜數字：與 Jev、Nimble 幾乎打平
在公開的 Decision Index 0.3.1（37 項 benchmark，經機率校正）上，Drex 自測拿到 58.08 分，是 10B 參數以下最高分，且與 Jev、Nimble 同在 0.9 分的誤差帶內，在 37 項 benchmark 中贏 Jev 20 項。細項上 Drex 的 Tools 類別最強（75.0），Knowledge 與 Reasoning 最弱（44.6）。在較舊的 Decision Index 0.2.1 上，Drex 拿 58.28 分、Jev 為 57.91 分。在 JevBench 的 231 道公開題目上，Drex 拿 86.2%、Jev 拿 87.0%，兩者在難題子集都是 73.9%。在 8 款 OpenSpiel 遊戲的正面對決中，Drex 以 122 勝 47 和 87 負（勝率 56.8%）贏過 Jev。長文件是 Drex 的強項：若把同樣的請求截斷到 8K tokens，準確率會掉到 76.5% 與 78%，顯示長上下文確實有加成。部署方面，Nace 在 AWS g5.2xlarge（A10G 24GB）上測試 bf16 與 Q8_0，答案一致；Q8_0 的 GGUF 版本在 Apple M5 Pro 的 Metal 與 CPU 上也給出相同結果。官方還提供 Drex agent skill，可接入 Claude Code、Codex、Cursor、OpenCode、Hermes Agent、Gemini CLI 與 GitHub Copilot。
⚠️ 知識類任務是明顯弱點
Drex 1.5 終究不是通用模型，無法生成文字、程式碼或任何解釋。在知識密集型測試上差距明顯：GPQA Diamond 只有 45.4%，對比 Jev 的 78.6%；MMLU-Pro 為 58.7%，對比 Jev 的 82.7%。由於訓練資料來自 benchmark 的官方訓練切分，模型在熟悉的決策類型上表現較好，換到全新領域的效果仍是未知數。本地部署若要用 Ollama 或 llama.cpp，需要 Nace 自己維護的 fork，而非主線版本。權重採用 RAIL-M 授權並附帶使用限制，商用前務必確認條款。
🎯 實務啟示
如果你的 agent 流程裡有大量「從固定選項中選一個」的環節（路由、審核、分類、遊戲決策），Drex 1.5 提供了一個開源且可本機部署的替代方案，API 相容於 Jev，遷移成本低；但知識類推理仍建議搭配一般 LLM 做第二層判斷。
🔗 來源
- 標題：Nace AI Open-Sources Drex 1.5: A 9B Decision Model That Scores Options, Not Text
- 作者／機構：Asif Razzaq, MarkTechPost
- 連結：https://www.marktechpost.com/2026/10/09/nace-ai-open-sources-drex-1-5-a-9b-decision-model-that-scores-options-not-text/
#DecisionModel #OpenSource #Qwen #AgentTools #LLM #HuggingFace #OpenRouter #MachineLearning #AIInfrastructure #NaceAI
原始資料 MarkTechPost · 收集於 2026-10-10
摘要原文
Nace.AI has open-sourced Drex 1.5 , a 9B decision model for agents and backend workflows. The Drex 1.5 decision model does not write text. It reads a state and typed questions, then returns a probability for every option. Nace reports 58.08 on the public Decision Index 0.3.1 , the top score under 10B parameters. Weights are on Hugging Face, and a hosted version is live on OpenRouter . Drex 1.5 is a decision model from Nace.AI that scores a fixed set of options in 1 forward pass. You send a state (text or JSON) and named questions. It supports 3 question types: choice , noul (yes/no) and ordinal score . No tokens are sampled, so temperature and top_p do not apply. The model can only answer with options you supplied. It serves the POST /v1/systemone API. That is the same request format used by TypeSafe’s Jev , the closed model that started this category. Nace says existing Jev clients work after changing a few environment variables. The backbone is MiMo-V2.6-Distill-Qwen-9B , a distilled Qwen 3.5 9B model. It has 32 layers with hybrid attention: 3 linear-attention layers per full-attention layer. A separate pointer head ( head.pt ) scores each option from the backbone’s hidden states. Each question runs 1 pass over the state plus that question. In llama.cpp, the state is encoded once and shared across questions. Nace’s Drex page says the model was trained on the official training splits of the index benchmarks. It was evaluated only on held-out splits. On the public Decision Index 0.3.1 (37 benchmarks, chance-corrected), the model card reports: Drex 1.5, Jev and Nimble sit within the board’s 0.9-point tie band. Drex leads Jev on 20 of 37 benchmarks. The Drex score comes from Nace’s own run of the official kit. Its area scores are strongest in Tools (75.0) and weakest in Knowledge and Reasoning (44.6). Nace’s launch chart uses the older Decision Index 0.2.1, where Drex scored 58.28 against Jev’s 57.91. On JevBench (231 public items), Drex scores 86.2% against Jev’s 87.0%. Both reach 73.9% on hard items. In a head-to-head across 8 OpenSpiel games, Drex recorded 122 wins, 47 draws and 87 losses against Jev (56.8%). Long documents are a clear strength: Truncating the same requests to 8K tokens drops accuracy to 76.5% and 78%. There are 4 deployment paths, all serving the same API: Nace tested bf16 and Q8_0 on an AWS g5.2xlarge (A10G 24 GB) and got identical answers. The Q8_0 GGUF also matched on an Apple M5 Pro on both Metal and CPU. A Drex agent skill plugs the model into Claude Code, Codex, Cursor, OpenCode, Hermes Agent, Gemini CLI and GitHub Copilot. Nace’s launch post also offers a $25 sign-up bonus for cloud users. Decision Index 0.3.1 rows are from the Drex 1.5 model card, which cites the public leaderboard for Jev and Nimble. Nimble’s 0.2.1 score is from its own model card. Drex 1.5 is a decision layer, not a general model. It cannot generate text, code or explanations. Knowledge-heavy tests are its weak spot: 45.4% on GPQA Diamond versus Jev’s 78.6%, and 58.7% on MMLU-Pro versus 82.7%. Training on the index benchmarks’ training splits helps it on familiar decision types. Results on new domains may differ. Local deployment through Ollama and llama.cpp needs Nace’s forks, not mainline builds. The weights use a RAIL-M license with use restrictions, so check the terms before commercial use. Check out the model weights on Hugging Face , the GitHub repo and the Drex product page . All credit goes to the researcher of this project. Also, feel free to follow us on Twitter and don’t forget to join our 150k+ML SubReddit and Subscribe to our Newsletter . Wait! are you on telegram? now you can join us on telegram as well. The post Nace AI Open-Sources Drex 1.5: A 9B Decision Model That Scores Options, Not Text appeared first on MarkTechPost .
由 claude-code/sonnet 自動生成