Prompt copied!
The text is now ready to paste anywhere.
Drag

Lightweight intelligence for enterprise applications

A 10B-parameter open-weight reasoning model for resource-constrained environments.

Technical Overview

Comprehensive Model Details

Model Specifications

Context Length
32K native tokens, extendable to 131K
License
MIT
Input/Output Type
Text
Languages
50+ languages
With emphasis on major European languages:  German, French, Italian, Spanish, Portuguese.

Model Performance

Model Performance

Domyn Small
Llama-3.1
405B
Microsoft phi-4
OpenAI gpt-5
xAI
Grok 4
DeepSeek
V3.1
ARC Challenge
94.8
94.8
94.8
94.8
94.8
94.8
MMLU
94.8
94.8
94.8
94.8
94.8
94.8
GPQA Diamond
94.8
94.8
94.8
94.8
94.8
94.8
Humanity’s Last Exam
94.8
94.8
94.8
94.8
94.8
94.8
AIME 2024
94.8
94.8
94.8
94.8
94.8
94.8
AIME 2024
94.8
94.8
94.8
94.8
94.8
94.8

Intended Use

Real-time interactive applications
Operates at the sub-second latencies required by voice assistants, live chat support, real-time autocomplete, and collaborative editing tools.
Resource-constrained deployment
Runs in scenarios where response times, cost, and computational resources are real constraints — including on laptops and mobile devices.
Agentic reasoning
Delivers reasoning efficiency for AI agents that solve complex coding and mathematical problems and perform sequential, multi-step tasks.
Custom model development
User-adaptable to industry-specific knowledge, terminology, and workflows through open-weight fine-tuning under the MIT license.

Features

10B parameters
Compact open-weight architecture delivering capable reasoning in resource-constrained environments.
Togglable reasoning mode
"Thinking on/off" toggle enabling deep multi-step reasoning or fast, compact output depending on the task.
32K tokens context window, 131K extended
Supports long-document analysis and multi-turn agent workflows thanks to 32K tokens native context window, extensible to 131K tokens via YaRN.
Multilingual European AI
Optimized for English and major European languages, with support for 50+ languages and a 256K-token multilingual vocabulary.
Maximum token efficiency
Fewer tokens generated than comparable models for equivalent reasoning tasks, lowering inference cost and latency in production.
Agent-ready architecture
Optimized for coding, math, instruction following, and multi-step agentic workflows.
Open-weight sovereign release
Enabling full ownership, on-premise deployment, and domain-specific fine-tuning.

Harnessing LLM
power across industries

From financial services to government and advanced manufacturing, Domyn Small can deliver actionable insights, optimize workflows, and provide strategic value.

Banking
Analyze transactions, regulatory filings, and market trends to optimize risk, detect anomalies, and support strategic banking decisions.
Insurance companies
Process claims, policies, and risk data to accelerate underwriting, detect fraud, and improve decision accuracy.
Utilities company
Interpret operational data, maintenance records, and regulatory information to optimize infrastructure management and ensure reliable service delivery.
Aerospace manufacturer
Analyze designs, production data, and technical documentation to improve efficiency, quality, and compliance.
Revenue agency
Process complex tax filings and compliance reports to streamline audits, flag inconsistencies, and enhance accuracy.
Department of education
Synthesize curricula, data, and insights to inform policy, planning, and resource allocation across educational programs.
Army
Analyze mission plans, logistics data, and operational intelligence to enhance coordination and tactical decision-making.
Air Force
Process operations data, maintenance logs, and intelligence reports to improve mission planning, airspace awareness, and fleet readiness.

Related Products

Our complete product offer

Explore our agents, platform, and compute layer to build a fully integrated AI stack — from chip to frontend.

it