凍結的12B模型在驗證任務上擊敗前沿模型：100%準確率、零Token、位元精確、永久有效

摘要

現今改善語言模型通常需要重新訓練：耗費大量運算、每個週期產生新的不透明模型、以及非確定性輸出。我們採取相反路徑：模型保持凍結，同時在模型旁增長一個關於已驗證解決方案的持久記憶體。一旦某問題家族被解決，並通過了從不查閱答案的獨立驗證步驟，該家族的每個新實例都能以零生成詞元、位元精確、確定性的方式得到回答。在橫跨九個問題家族、來自四個供應商的四種架構（密集型與專家混合型）的180個新實例中，每個模型在每次回答使用零生成詞元的條件下均獲得180/180分：執行綁定能力已與參數縮放脫鉤。一項陰性對照將能力完全歸因於記憶體：一旦清空記憶體，系統便無法解決任何問題。同樣的「驗證再儲存」契約也適用於開放式推理：所有四個模型均達到88/88的一致性門控接受、機器可檢查的形式化證明，以及77/80的推理方法遷移率。記憶體選取僅需1.4微秒；完整重用則在6至23毫秒內完成，功耗36毫瓦時。在一個包含4,500項驗證記憶體的儲存中，近似相似性檢索在精確定址零錯誤的情況下，有94.3%的機率選錯項目。該記憶體還能作為工作上下文使用，其規模是任何已發布引擎都無法比擬的：在單一46 GB GPU且記憶體平穩的條件下，可提供600萬詞元的可移動視窗，而vLLM在30,399詞元處即停止，SGLang則在超過32,000詞元時默默截斷。在已發表的基準測試中，前沿模型在原始從頭推理上仍遠勝任何12B模型；但針對本系統已解決並驗證過的所有問題，比較結果出現逆轉：前沿API呼叫每次查詢都需支付全新的生成過程代價，永不停止；而驗證後重用則耗費零詞元，且每次返回完全相同的位元。本報告附帶一個提供免費、限速存取的公開測試平台：https://corbenic-galahad-bench.hf.space

English

Improving a language model today means retraining it: enormous compute, a new opaque model each cycle, non-deterministic output. We take the opposite path: the model stays frozen, and a persistent memory of verified solutions grows beside it. Once a problem family is solved and has passed an independent verification step that never consults the answer key, every new instance of that family is answered at zero generation tokens, bit-exact, deterministically. Across 180 fresh instances spanning nine problem families, four architectures from four vendors - dense and mixture-of-experts - each score 180/180 at zero generation tokens per answer: execution-bound capability decoupled from parameter scaling. A negative control attributes the capability fully to the memory: emptied, it solves nothing. The same verify-before-store contract holds for open-ended reasoning: 88/88 consistency-gated acceptances across all four models, machine-checked formal proof, and reasoning-method transfer at 77/80. Memory selection takes 1.4 microseconds; a full reuse completes in 6-23 ms at 36 mWh. Approximate similarity retrieval selects the wrong item 94.3% of the time on a 4,500-item verified store where exact addressing makes zero errors. The store also serves as working context at a scale no shipped engine matches: a 6,000,000-token movable window on a single 46 GB GPU at flat memory, where vLLM stops at 30,399 tokens and SGLang silently truncates past 32,000. On published benchmarks, frontier models remain far ahead of any 12B at raw from-scratch reasoning; on everything this system has solved and verified, the comparison inverts: a frontier API call pays a fresh generation pass on every query, forever, while verified reuse costs zero tokens and returns the identical bits every time. A public testbench with free, rate-limited access accompanies this report: https://corbenic-galahad-bench.hf.space