{"id":284,"date":"2026-11-02T08:22:02","date_gmt":"2026-11-02T02:52:02","guid":{"rendered":"https:\/\/akshaymaurya.in\/blog\/?p=284"},"modified":"2026-11-02T08:22:04","modified_gmt":"2026-11-02T02:52:04","slug":"nace-ai-drex-max-v1-open-weight-hindi","status":"publish","type":"post","link":"https:\/\/akshaymaurya.in\/blog\/nace-ai-drex-max-v1-open-weight-hindi\/","title":{"rendered":"Nace AI Drex Max v1: 27B Parameter \u0935\u093e\u0932\u093e Open-Weight Decision Model \u091c\u094b Text \u0928\u0939\u0940\u0902 \u0932\u093f\u0916\u0924\u093e, \u092b\u0948\u0938\u0932\u0947 \u0932\u0947\u0924\u093e \u0939\u0948"},"content":{"rendered":"\n

1 November 2026 \u0915\u094b Nace AI \u0928\u0947 \u0905\u092a\u0928\u0947 \u0905\u092c \u0924\u0915 \u0915\u0947 \u0938\u092c\u0938\u0947 capable model Drex Max v1 \u0915\u0947 \u092a\u0942\u0930\u0947 weights public \u0915\u0930 \u0926\u093f\u090f\u0964 \u0906\u0938\u093e\u0928 \u0936\u092c\u094d\u0926\u094b\u0902 \u092e\u0947\u0902 \u0915\u0939\u0947\u0902 \u0924\u094b \u092f\u0939 \u0910\u0938\u093e AI model \u0939\u0948 \u091c\u094b \u0915\u0939\u093e\u0928\u0940, \u0915\u0935\u093f\u0924\u093e \u092f\u093e \u091c\u0935\u093e\u092c \u0932\u093f\u0916\u0924\u093e \u0928\u0939\u0940\u0902 \u0939\u0948\u0964 \u0907\u0938\u0947 \u0906\u092a \u0915\u094b\u0908 situation \u0926\u0940\u091c\u093f\u090f \u0914\u0930 \u0915\u0941\u091b options \u0926\u0940\u091c\u093f\u090f, \u0924\u094b \u092f\u0939 \u0939\u0930 option \u0915\u0947 \u0932\u093f\u090f probability \u092c\u0924\u093e \u0926\u0947\u0917\u093e\u0964 \u092c\u0938\u0964 \u090f\u0915 forward pass \u092e\u0947\u0902 \u092b\u0948\u0938\u0932\u093e, \u092c\u093f\u0928\u093e word generate \u0915\u093f\u090f\u0964<\/p>\n

Drex Max \u0915\u094d\u092f\u093e \u0939\u0948<\/h2>\n

Drex Max \u0915\u094b Nace AI<\/a> \u0928\u0947 decisions \u0915\u0947 \u0932\u093f\u090f \u092c\u0928\u093e\u092f\u093e \u0939\u0948\u0964 Normal AI models \u0938\u0935\u093e\u0932 \u0915\u093e \u091c\u0935\u093e\u092c text \u092e\u0947\u0902 \u0926\u0947\u0924\u0947 \u0939\u0948\u0902\u0964 Drex Max \u0909\u0932\u094d\u091f\u093e \u0915\u093e\u092e \u0915\u0930\u0924\u093e \u0939\u0948\u0964 \u0906\u092a \u0909\u0938\u0947 \u090f\u0915 state \u0926\u0947\u0924\u0947 \u0939\u0948\u0902, \u092f\u093e\u0928\u0940 context, \u091c\u094b text, JSON \u092f\u093e images \u0939\u094b \u0938\u0915\u0924\u093e \u0939\u0948\u0964 \u092b\u093f\u0930 \u0906\u092a \u0909\u0938\u0938\u0947 typed questions \u092a\u0942\u091b\u0924\u0947 \u0939\u0948\u0902\u0964 Questions \u0924\u0940\u0928 \u0924\u0930\u0939 \u0915\u0947 \u0939\u094b\u0924\u0947 \u0939\u0948\u0902\u0964 \u092a\u0939\u0932\u093e, choice, \u092f\u093e\u0928\u0940 \u0926\u093f\u090f \u0917\u090f options \u092e\u0947\u0902 \u0938\u0947 \u091a\u0941\u0928\u0928\u093e\u0964 \u0926\u0942\u0938\u0930\u093e, yes \u092f\u093e no\u0964 \u0924\u0940\u0938\u0930\u093e, ordinal score, \u092f\u093e\u0928\u0940 \u0915\u093f\u0938\u0940 scale \u092a\u0930 rating \u0926\u0947\u0928\u093e\u0964<\/p>\n

Model \u092b\u093f\u0930 \u0939\u0930 option \u0915\u0947 \u0932\u093f\u090f \u090f\u0915 probability return \u0915\u0930\u0924\u093e \u0939\u0948\u0964 \u0909\u0926\u093e\u0939\u0930\u0923 \u0915\u0947 \u0932\u093f\u090f, email spam \u0939\u0948 \u092f\u093e \u0928\u0939\u0940\u0902, \u0907\u0938 \u092a\u0930 model \u091c\u0935\u093e\u092c \u0926\u0947\u0917\u093e: spam 92%, not spam 8%\u0964 \u0915\u094b\u0908 explanation \u0928\u0939\u0940\u0902, \u0938\u093f\u0930\u094d\u092b numbers\u0964 \u092e\u0947\u0930\u0947 \u0939\u093f\u0938\u093e\u092c \u0938\u0947 \u092f\u0939\u0940 \u0907\u0938\u0915\u0940 \u0938\u092c\u0938\u0947 \u092c\u0921\u093c\u0940 \u0916\u0942\u092c\u0940 \u0939\u0948\u0964 AI agents \u0915\u094b \u092b\u0948\u0938\u0932\u0947 \u0915\u0947 \u0932\u093f\u090f \u0915\u0939\u093e\u0928\u0940 \u0928\u0939\u0940\u0902, numbers \u091a\u093e\u0939\u093f\u090f \u0939\u094b\u0924\u0947 \u0939\u0948\u0902\u0964<\/p>\n

\u092f\u0939 model \u0915\u093e\u092e \u0915\u0948\u0938\u0947 \u0915\u0930\u0924\u093e \u0939\u0948<\/h2>\n

Architecture \u0915\u0940 \u092c\u093e\u0924 \u0915\u0930\u0947\u0902 \u0924\u094b Drex Max \u090f\u0915 dense 27 billion parameter \u0935\u093e\u0932\u093e model \u0939\u0948\u0964 \u0907\u0938\u0915\u0947 64 layers \u092e\u0947\u0902 \u0938\u0947 48 Gated DeltaNet linear-attention \u0935\u093e\u0932\u0947 \u0939\u0948\u0902\u0964 \u092c\u093e\u0915\u0940 16 full-attention layers \u0939\u0930 \u091a\u094c\u0925\u0947 layer \u092a\u0930 \u0939\u0948\u0902 \u0914\u0930 \u092a\u0942\u0930\u0947 sequence \u0915\u094b \u090f\u0915 \u0938\u093e\u0925 \u0926\u0947\u0916 \u0938\u0915\u0924\u0947 \u0939\u0948\u0902\u0964<\/p>\n

Context window 262,144 tokens \u0915\u0940 \u0939\u0948, default limit 16,384 tokens \u0930\u0916\u0940 \u0917\u0908 \u0939\u0948\u0964 \u090f\u0915 \u0938\u0935\u093e\u0932 \u092e\u0947\u0902 255 \u0924\u0915 options \u0926\u093f\u090f \u091c\u093e \u0938\u0915\u0924\u0947 \u0939\u0948\u0902, \u0915\u094d\u092f\u094b\u0902\u0915\u093f model \u0915\u0947 \u0905\u0902\u0926\u0930 255-way readout head \u0939\u0948 \u091c\u094b last hidden state \u0915\u094b option probabilities \u092e\u0947\u0902 \u092c\u0926\u0932 \u0926\u0947\u0924\u093e \u0939\u0948\u0964<\/p>\n

\u0915\u0939\u093e\u0902 \u0907\u0938\u094d\u0924\u0947\u092e\u093e\u0932 \u0939\u094b\u0917\u093e \u092f\u0939 model<\/h2>\n

Nace AI \u0907\u0938\u0947 \u0924\u0940\u0928 \u0915\u093e\u092e\u094b\u0902 \u0915\u0947 \u0932\u093f\u090f position \u0915\u0930 \u0930\u0939\u093e \u0939\u0948\u0964 \u092a\u0939\u0932\u093e, agent routing, \u092f\u093e\u0928\u0940 AI agent \u0915\u094b \u092f\u0939 \u0924\u092f \u0915\u0930\u0928\u0947 \u092e\u0947\u0902 \u092e\u0926\u0926 \u0915\u093f \u0905\u0917\u0932\u093e step \u0915\u094d\u092f\u093e \u0939\u094b\u0964 \u0926\u0942\u0938\u0930\u093e, ticket triage, \u092f\u093e\u0928\u0940 customer support tickets \u0915\u094b \u0938\u0939\u0940 category \u092e\u0947\u0902 \u092c\u093e\u0902\u091f\u0928\u093e\u0964 \u0924\u0940\u0938\u0930\u093e, policy checks \u0914\u0930 guardrails, \u092f\u093e\u0928\u0940 \u092f\u0939 \u091c\u093e\u0902\u091a\u0928\u093e \u0915\u093f \u0915\u094b\u0908 content \u092f\u093e action rules \u0915\u0947 \u0916\u093f\u0932\u093e\u092b \u0924\u094b \u0928\u0939\u0940\u0902 \u0939\u0948\u0964<\/p>\n

\u091c\u094b \u0932\u094b\u0917 \u0916\u0941\u0926 model \u0928\u0939\u0940\u0902 \u091a\u0932\u093e\u0928\u093e \u091a\u093e\u0939\u0924\u0947, \u0909\u0928\u0915\u0947 \u0932\u093f\u090f hosted API \u0939\u0948\u0964 OpenRouter \u092a\u0930 nace-ai\/drex-max \u0928\u093e\u092e \u0938\u0947 \u092f\u0939 model $0.06 per 1M input tokens \u092a\u0930 \u092e\u093f\u0932\u0924\u093e \u0939\u0948, \u0914\u0930 output tokens \u092a\u0930 \u0915\u094b\u0908 charge \u0928\u0939\u0940\u0902 \u0939\u0948, \u0915\u094d\u092f\u094b\u0902\u0915\u093f output \u092e\u0947\u0902 text generate \u0939\u094b\u0924\u093e \u0939\u0940 \u0928\u0939\u0940\u0902\u0964<\/p>\n

\u0905\u092a\u0928\u0947 computer \u092a\u0930 \u0915\u0948\u0938\u0947 \u091a\u0932\u093e\u090f\u0902<\/h2>\n

Open weights release \u0915\u093e \u0938\u092c\u0938\u0947 \u092c\u0921\u093c\u093e \u092b\u093e\u092f\u0926\u093e \u092f\u0939\u0940 \u0939\u0948 \u0915\u093f model \u0915\u094b \u0905\u092a\u0928\u0940 machine \u092a\u0930 \u091a\u0932\u093e\u092f\u093e \u091c\u093e \u0938\u0915\u0924\u093e \u0939\u0948, data \u092c\u093e\u0939\u0930 \u092d\u0947\u091c\u0947 \u092c\u093f\u0928\u093e\u0964 Release \u092e\u0947\u0902 bf16 weights (\u0915\u0930\u0940\u092c 49 GB), Q8_0 GGUF (\u0915\u0930\u0940\u092c 27.25 GB) \u0914\u0930 affine Q4 checkpoint \u0926\u093f\u090f \u0917\u090f \u0939\u0948\u0902\u0964 \u0938\u093e\u0925 \u092e\u0947\u0902 llama.cpp \u0914\u0930 Ollama \u0915\u0947 forks \u092d\u0940 \u0939\u0948\u0902\u0964<\/p>\n

Hardware \u0915\u0940 \u091c\u0930\u0942\u0930\u0924 \u0907\u0938 \u092a\u094d\u0930\u0915\u093e\u0930 \u0939\u0948\u0964 NVIDIA GPU \u092a\u0930 bf16 \u0915\u0947 \u0932\u093f\u090f H100 \u091c\u0948\u0938\u093e card \u091a\u093e\u0939\u093f\u090f, \u0914\u0930 8-bit \u0915\u0947 \u0932\u093f\u090f 48 GB \u0935\u093e\u0932\u093e L40S \u0915\u093e\u092b\u0940 \u0939\u0948\u0964 Apple Silicon Macs \u092a\u0930 24 GB \u092f\u093e \u091c\u094d\u092f\u093e\u0926\u093e memory \u0939\u094b \u0924\u094b affine Q4 checkpoint Kapitsa \u092f\u093e MLX \u0938\u0947 \u091a\u0932\u0924\u093e \u0939\u0948\u0964 48 GB memory \u0935\u093e\u0932\u0947 Mac \u092a\u0930 Q8_0 GGUF \u092d\u0940 \u091a\u0932\u0947\u0917\u093e\u0964 CPU \u092a\u0930 llama.cpp \u0914\u0930 Ollama \u0915\u0947 \u091c\u0930\u093f\u090f \u092d\u0940 \u0915\u093e\u092e \u0939\u094b \u0938\u0915\u0924\u093e \u0939\u0948\u0964<\/p>\n

Benchmark numbers: company \u0915\u0947 \u0905\u092a\u0928\u0947 \u0926\u093e\u0935\u0947<\/h2>\n

Benchmarks \u0915\u094b \u0939\u092e\u0947\u0936\u093e company \u0915\u0947 \u0926\u093e\u0935\u0947 \u0915\u0940 \u0924\u0930\u0939 \u092a\u0922\u093c\u0928\u093e \u091a\u093e\u0939\u093f\u090f\u0964 Nace AI \u0915\u0947 \u092e\u0941\u0924\u093e\u092c\u093f\u0915 Drex Max \u0928\u0947 public Decision Index 0.4 \u092a\u0930 67.52 score \u0915\u093f\u092f\u093e \u0939\u0948\u0964 \u0924\u0941\u0932\u0928\u093e \u0915\u0947 \u0932\u093f\u090f Drex v2.0 \u0915\u093e score 61.25 \u0914\u0930 TypeSafe \u0915\u0947 Jev 1.13.0 \u0915\u093e 57.96 \u0939\u0948\u0964 \u092f\u093e\u0928\u0940 \u0905\u092a\u0928\u0947 \u092a\u093f\u091b\u0932\u0947 version \u0938\u0947 6.27 points \u0915\u0940 \u092c\u0922\u093c\u0924\u0964<\/p>\n

Public vision index \u092a\u0930 score 73.92 \u0939\u0948\u0964 Artificial Analysis \u0915\u0947 Decision Index \u092a\u0930 score 64 \u0939\u0948, \u091c\u094b 18 models \u092e\u0947\u0902 4th rank \u0926\u0947\u0924\u093e \u0939\u0948\u0964 InfographicVQA \u092a\u0930 97.3 skill score \u0914\u0930 Tools area \u092e\u0947\u0902 81.9 score \u092c\u0924\u093e\u092f\u093e \u0917\u092f\u093e \u0939\u0948\u0964 Test \u092e\u0947\u0902 140,620 requests \u0925\u0947, \u0914\u0930 unanswered requests \u0915\u094b \u0917\u0932\u0924 \u092e\u093e\u0928\u093e \u0917\u092f\u093e\u0964 \u092e\u0947\u0930\u0947 \u0939\u093f\u0938\u093e\u092c \u0938\u0947 \u092f\u0939 honest evaluation \u0939\u0948, \u0915\u094d\u092f\u094b\u0902\u0915\u093f \u092c\u0939\u0941\u0924 \u0938\u0940 companies unanswered requests \u0915\u094b \u0917\u093f\u0928\u0924\u0940 \u0938\u0947 \u092c\u093e\u0939\u0930 \u0915\u0930 \u0926\u0947\u0924\u0940 \u0939\u0948\u0902\u0964<\/p>\n

License \u0915\u0940 \u092c\u093e\u0930\u0940\u0915\u093f\u092f\u093e\u0902 \u0938\u092e\u091d \u0932\u0940\u091c\u093f\u090f<\/h2>\n

Weights Nace.AI Open RAIL-M license \u0915\u0947 \u0924\u0939\u0924 release \u0939\u0941\u090f \u0939\u0948\u0902\u0964 \u092f\u0939\u093e\u0902 \u090f\u0915 \u092c\u093e\u0924 \u0938\u093e\u092b \u0938\u092e\u091d \u0932\u0940\u091c\u093f\u090f\u0964 Open-weight \u0915\u093e \u092e\u0924\u0932\u092c open-source \u0928\u0939\u0940\u0902 \u0939\u094b\u0924\u093e\u0964 \u0906\u092a\u0915\u094b model \u0915\u093e \u0938\u0940\u0916\u093e \u0939\u0941\u0906 \u091c\u094d\u091e\u093e\u0928, \u092f\u093e\u0928\u0940 weights \u092e\u093f\u0932 \u0930\u0939\u0947 \u0939\u0948\u0902\u0964 Training data \u0914\u0930 training code company \u0928\u0947 share \u0928\u0939\u0940\u0902 \u0915\u093f\u090f \u0939\u0948\u0902\u0964 \u092f\u0939 \u0935\u0939\u0940 \u092b\u0930\u094d\u0915 \u0939\u0948 \u091c\u094b \u0939\u092e\u0928\u0947 Mistral Large 4 'Le Chonk'<\/a> \u0915\u0947 launch \u092a\u0930 \u092d\u0940 \u0926\u0947\u0916\u093e \u0925\u093e\u0964 \u0935\u0939\u093e\u0902 \u092d\u0940 weights \u0916\u0941\u0932\u0947 \u0925\u0947, recipe \u0928\u0939\u0940\u0902\u0964<\/p>\n

RAIL-M \u092e\u0947\u0902 responsible use \u0915\u0940 \u0936\u0930\u094d\u0924\u0947\u0902 \u0939\u0948\u0902\u0964 Production \u092e\u0947\u0902 \u0909\u0924\u093e\u0930\u0928\u0947 \u0938\u0947 \u092a\u0939\u0932\u0947 license \u091c\u0930\u0942\u0930 \u092a\u0922\u093c\u0947\u0902\u0964<\/p>\n

\u092e\u0941\u0915\u093e\u092c\u0932\u093e: Mistral Large 4 \u0914\u0930 Jev \u0938\u0947 \u0915\u0939\u093e\u0902 \u0916\u0921\u093c\u093e \u0939\u0948<\/h2>\n

Decision models \u0915\u0940 \u0926\u0941\u0928\u093f\u092f\u093e \u092e\u0947\u0902 Drex Max \u0915\u093e \u0938\u0940\u0927\u093e \u092e\u0941\u0915\u093e\u092c\u0932\u093e TypeSafe AI \u0915\u0947 Jev models \u0938\u0947 \u0939\u0948\u0964 Numbers \u0915\u0947 \u0939\u093f\u0938\u093e\u092c \u0938\u0947 Drex Max Jev 1.13.0 \u0938\u0947 \u0906\u0917\u0947 \u0939\u0948\u0964 \u0932\u0947\u0915\u093f\u0928 \u090f\u0915 \u0914\u0930 \u0924\u0941\u0932\u0928\u093e \u0926\u093f\u0932\u091a\u0938\u094d\u092a \u0939\u0948, Mistral Large 4 \u0938\u0947\u0964 Mistral \u0915\u093e model 1 trillion parameter \u0935\u093e\u0932\u093e generative giant \u0939\u0948, \u091c\u092c\u0915\u093f Drex Max 27B \u0915\u093e decision specialist\u0964 \u0926\u094b\u0928\u094b\u0902 \u0915\u0940 race \u0905\u0932\u0917 \u0939\u0948\u0964 \u090f\u0915 \u0915\u0939\u093e\u0928\u0940 \u0932\u093f\u0916\u0924\u093e \u0939\u0948, \u0926\u0942\u0938\u0930\u093e \u092b\u0948\u0938\u0932\u093e \u0938\u0941\u0928\u093e\u0924\u093e \u0939\u0948\u0964<\/p>\n

\u092e\u0947\u0930\u0947 \u0939\u093f\u0938\u093e\u092c \u0938\u0947 Drex Max \u0915\u0940 \u0905\u0938\u0932\u0940 \u0915\u0939\u093e\u0928\u0940 category \u092e\u0947\u0902 \u0939\u0948\u0964 Generative models \u0915\u0940 race \u092e\u0947\u0902 \u0938\u092c \u0932\u0917\u0947 \u0939\u0948\u0902, \u0932\u0947\u0915\u093f\u0928 agents \u0915\u0947 \u0932\u093f\u090f \u091c\u0930\u0942\u0930\u0940 decision models \u092e\u0947\u0902 competition \u0915\u092e \u0939\u0948\u0964 Nace AI \u0928\u0947 \u0909\u0938\u0940 gap \u0915\u094b \u092a\u0915\u0921\u093c\u093e \u0939\u0948, \u0914\u0930 weights \u0916\u094b\u0932\u0915\u0930 developers \u0915\u093e \u092d\u0930\u094b\u0938\u093e \u091c\u0940\u0924\u0928\u0947 \u0915\u0940 \u0915\u094b\u0936\u093f\u0936 \u0915\u0940 \u0939\u0948\u0964<\/p>\n

\u0906\u092e user \u0915\u0947 \u0932\u093f\u090f \u0907\u0938\u0915\u093e \u092e\u0924\u0932\u092c \u0915\u094d\u092f\u093e \u0939\u0948<\/h2>\n

\u0938\u0940\u0927\u0940 \u092c\u093e\u0924, \u0906\u092e user \u0907\u0938\u0947 download \u0915\u0930\u0915\u0947 \u0928\u0939\u0940\u0902 \u091a\u0932\u093e\u090f\u0917\u093e\u0964 27B model \u0906\u092e laptop \u0915\u0947 \u092c\u0938 \u0915\u0940 \u092c\u093e\u0924 \u0928\u0939\u0940\u0902, \u0914\u0930 \u092f\u0939 chat \u0915\u0947 \u0932\u093f\u090f \u092c\u0928\u093e \u092d\u0940 \u0928\u0939\u0940\u0902 \u0939\u0948\u0964 \u0932\u0947\u0915\u093f\u0928 \u092b\u093e\u092f\u0926\u093e indirect \u0924\u0930\u0940\u0915\u0947 \u0938\u0947 \u092a\u0939\u0941\u0902\u091a\u0947\u0917\u093e\u0964<\/p>\n

\u092a\u0939\u0932\u093e, \u091c\u092c developers \u0910\u0938\u0947 decision models \u0915\u094b \u0905\u092a\u0928\u0947 server \u092a\u0930 \u091a\u0932\u093e \u0938\u0915\u0947\u0902\u0917\u0947, \u0924\u094b AI apps \u092e\u0947\u0902 routing \u0914\u0930 safety checks \u0938\u0938\u094d\u0924\u0947 \u0939\u094b\u0902\u0917\u0947\u0964 \u0926\u0942\u0938\u0930\u093e, privacy\u0964 Policy checks \u0905\u0917\u0930 local machine \u092a\u0930 \u0939\u094b\u0902, \u0924\u094b \u0906\u092a\u0915\u093e data cloud \u0924\u0915 \u091c\u093e\u0928\u0947 \u0915\u0940 \u091c\u0930\u0942\u0930\u0924 \u0928\u0939\u0940\u0902\u0964 \u0924\u0940\u0938\u0930\u093e, \u091b\u094b\u091f\u0947 startups \u0905\u092c \u092c\u093f\u0928\u093e \u092e\u0939\u0902\u0917\u0940 API \u0915\u0947 \u0905\u092a\u0928\u0947 agents \u092e\u0947\u0902 decision layer \u091c\u094b\u0921\u093c \u0938\u0915\u0947\u0902\u0917\u0947\u0964 \u092e\u0947\u0930\u0947 \u0939\u093f\u0938\u093e\u092c \u0938\u0947 \u092f\u0939 launch generative AI \u0915\u0940 \u091a\u092e\u0915 \u0938\u0947 \u0926\u0942\u0930, \u0932\u0947\u0915\u093f\u0928 agentic AI \u0915\u0940 \u0928\u0940\u0902\u0935 \u0915\u094b \u092e\u091c\u092c\u0942\u0924 \u0915\u0930\u0928\u0947 \u0935\u093e\u0932\u093e \u0915\u0926\u092e \u0939\u0948\u0964<\/p>\n

FAQ<\/h2>\n

Drex Max v1 \u0915\u094d\u092f\u093e \u0939\u0948<\/h3>\n

\u092f\u0939 Nace AI \u0915\u093e 27 billion parameter \u0935\u093e\u0932\u093e decision model \u0939\u0948, \u091c\u093f\u0938\u0915\u0947 weights 1 November 2026 \u0915\u094b open \u0915\u093f\u090f \u0917\u090f\u0964 \u092f\u0939 text generate \u0928\u0939\u0940\u0902 \u0915\u0930\u0924\u093e, \u0926\u093f\u090f \u0917\u090f options \u0915\u0947 \u0932\u093f\u090f probability \u0926\u0947\u0924\u093e \u0939\u0948\u0964<\/p>\n

Decision model \u0914\u0930 normal AI model \u092e\u0947\u0902 \u0915\u094d\u092f\u093e \u092b\u0930\u094d\u0915 \u0939\u0948<\/h3>\n

Normal AI model text \u092e\u0947\u0902 \u091c\u0935\u093e\u092c \u0926\u0947\u0924\u093e \u0939\u0948\u0964 Decision model \u0938\u0935\u093e\u0932 \u0914\u0930 options \u0932\u0947\u0915\u0930 \u0939\u0930 option \u0915\u0940 probability \u090f\u0915 forward pass \u092e\u0947\u0902 return \u0915\u0930\u0924\u093e \u0939\u0948\u0964<\/p>\n

\u0915\u094d\u092f\u093e \u0907\u0938\u0947 free \u092e\u0947\u0902 use \u0915\u0930 \u0938\u0915\u0924\u0947 \u0939\u0948\u0902<\/h3>\n

Weights Nace.AI Open RAIL-M license \u0915\u0947 \u0924\u0939\u0924 free download \u0915\u0947 \u0932\u093f\u090f \u0939\u0948\u0902\u0964 Hosted API OpenRouter \u092a\u0930 $0.06 per 1M input tokens \u092a\u0930 \u092e\u093f\u0932\u0924\u0940 \u0939\u0948\u0964<\/p>\n

\u0907\u0938\u0947 \u091a\u0932\u093e\u0928\u0947 \u0915\u0947 \u0932\u093f\u090f \u0915\u094d\u092f\u093e hardware \u091a\u093e\u0939\u093f\u090f<\/h3>\n

NVIDIA GPU \u092a\u0930 H100 \u092f\u093e 48 GB \u0935\u093e\u0932\u093e L40S\u0964 Apple Silicon Mac \u092a\u0930 24 GB \u092f\u093e \u091c\u094d\u092f\u093e\u0926\u093e memory\u0964 CPU \u092a\u0930 llama.cpp \u0914\u0930 Ollama \u0938\u0947 \u092d\u0940 \u091a\u0932 \u0938\u0915\u0924\u093e \u0939\u0948\u0964<\/p>\n

\u0907\u0938\u0915\u0947 benchmark scores \u0915\u094d\u092f\u093e \u0939\u0948\u0902<\/h3>\n

Company \u0915\u0947 \u0926\u093e\u0935\u094b\u0902 \u0915\u0947 \u092e\u0941\u0924\u093e\u092c\u093f\u0915 Decision Index 0.4 \u092a\u0930 67.52, \u091c\u094b Drex v2.0 \u0915\u0947 61.25 \u0914\u0930 Jev 1.13.0 \u0915\u0947 57.96 \u0938\u0947 \u091c\u094d\u092f\u093e\u0926\u093e \u0939\u0948\u0964<\/p>\n

\u0915\u094d\u092f\u093e \u092f\u0939 \u092a\u0942\u0930\u0940 \u0924\u0930\u0939 open-source \u0939\u0948<\/h3>\n

\u0928\u0939\u0940\u0902\u0964 \u092f\u0939 open-weight release \u0939\u0948\u0964 Weights public \u0939\u0948\u0902, \u0932\u0947\u0915\u093f\u0928 training data \u0914\u0930 training code share \u0928\u0939\u0940\u0902 \u0915\u093f\u090f \u0917\u090f \u0939\u0948\u0902\u0964<\/p>\n","protected":false},"excerpt":{"rendered":"

1 November 2026 \u0915\u094b Nace AI \u0928\u0947 \u0905\u092a\u0928\u0947 \u0905\u092c \u0924\u0915 \u0915\u0947 \u0938\u092c\u0938\u0947 capable model Drex Max v1 \u0915\u0947 \u092a\u0942\u0930\u0947 weights public \u0915\u0930 \u0926\u093f\u090f\u0964 \u0906\u0938\u093e\u0928 \u0936\u092c\u094d\u0926\u094b\u0902 \u092e\u0947\u0902 \u0915\u0939\u0947\u0902 \u0924\u094b \u092f\u0939 \u0910\u0938\u093e AI model \u0939\u0948 \u091c\u094b \u0915\u0939\u093e\u0928\u0940, \u0915\u0935\u093f\u0924\u093e \u092f\u093e \u091c\u0935\u093e\u092c \u0932\u093f\u0916\u0924\u093e \u0928\u0939\u0940\u0902 \u0939\u0948\u0964 \u0907\u0938\u0947 \u0906\u092a \u0915\u094b\u0908 situation \u0926\u0940\u091c\u093f\u090f \u0914\u0930 \u0915\u0941\u091b options \u0926\u0940\u091c\u093f\u090f, \u0924\u094b \u092f\u0939 \u0939\u0930 option \u0915\u0947 […]<\/p>\n","protected":false},"author":1,"featured_media":277,"comment_status":"open","ping_status":"open","sticky":false,"template":"","format":"standard","meta":{"_jetpack_newsletter_access":"","_jetpack_dont_email_post_to_subs":false,"_jetpack_newsletter_tier_id":0,"_jetpack_memberships_contains_paywalled_content":false,"_jetpack_hide_featured_image":false,"_jetpack_featured_image_duplicate_dismissed":0,"_jetpack_feature_clip_id":0,"_jetpack_memberships_contains_paid_content":false,"footnotes":"","jetpack_post_was_ever_published":false},"categories":[1],"tags":[100,99,96,97,98],"class_list":["post-284","post","type-post","status-publish","format-standard","has-post-thumbnail","hentry","category-artificial-intelligence","tag-ai-benchmarks","tag-decision-model","tag-drex-max","tag-nace-ai","tag-open-weight-model"],"jetpack_sharing_enabled":true,"jetpack_featured_media_url":"https:\/\/akshaymaurya.in\/blog\/wp-content\/uploads\/2026\/11\/thumb-nace-ai-drex-max-v1-open-weight-hindi.jpg","_links":{"self":[{"href":"https:\/\/akshaymaurya.in\/blog\/wp-json\/wp\/v2\/posts\/284","targetHints":{"allow":["GET"]}}],"collection":[{"href":"https:\/\/akshaymaurya.in\/blog\/wp-json\/wp\/v2\/posts"}],"about":[{"href":"https:\/\/akshaymaurya.in\/blog\/wp-json\/wp\/v2\/types\/post"}],"author":[{"embeddable":true,"href":"https:\/\/akshaymaurya.in\/blog\/wp-json\/wp\/v2\/users\/1"}],"replies":[{"embeddable":true,"href":"https:\/\/akshaymaurya.in\/blog\/wp-json\/wp\/v2\/comments?post=284"}],"version-history":[{"count":1,"href":"https:\/\/akshaymaurya.in\/blog\/wp-json\/wp\/v2\/posts\/284\/revisions"}],"predecessor-version":[{"id":285,"href":"https:\/\/akshaymaurya.in\/blog\/wp-json\/wp\/v2\/posts\/284\/revisions\/285"}],"wp:featuredmedia":[{"embeddable":true,"href":"https:\/\/akshaymaurya.in\/blog\/wp-json\/wp\/v2\/media\/277"}],"wp:attachment":[{"href":"https:\/\/akshaymaurya.in\/blog\/wp-json\/wp\/v2\/media?parent=284"}],"wp:term":[{"taxonomy":"category","embeddable":true,"href":"https:\/\/akshaymaurya.in\/blog\/wp-json\/wp\/v2\/categories?post=284"},{"taxonomy":"post_tag","embeddable":true,"href":"https:\/\/akshaymaurya.in\/blog\/wp-json\/wp\/v2\/tags?post=284"}],"curies":[{"name":"wp","href":"https:\/\/api.w.org\/{rel}","templated":true}]}}