Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Built upon extensive training, Qwen3 delivers groundbreaking advancements in reasoning, instruction-following, agent capabilities, and multilingual support, with the following key features: - Uniquely support of seamless switching between thinking mode (for complex logical reasoning, math, and coding) and non-thinking mode (for efficient, general-purpose dialogue) within single model, ensuring optimal performance across various scenarios. - Significantly enhancement in its reasoning capabilities, surpassing previous QwQ (in thinking mode) and…
Open-weight model · Text generation
Keural-Cortex-8B-v1.0-1
by Hika Dalju Hikaf143/Keural-Cortex-8B-v1.0-1
Keural-Cortex-8B-v1.0-1 is an open-weight model for text generation from Hika Dalju, released under Apache License 2.0. It has 8.2B parameters and a 65,536-token context. At 16-bit it needs about 19.7 GB of GPU memory, which fits on 1x MI300X from $1.85 an hour, at the lowest prices in the SAVRN Index.
Keural Cortex is MKD's local-first Korean-English language-model program for Keural applications. This release is an 8B-parameter, instruction- and preference-aligned checkpoint with a structured tool-calling layer for enterprise workflows.
Parameters8.2B
Context65,536
Weights16.4 GB
Licenseapache-2.0
AccessOpen weights
Monthly Downloads—
Runs On
What it takes to serve Keural-Cortex-8B-v1.0-1 (8.2B parameters): the memory its weights need at each precision, and the cheapest way to rent enough data-center GPUs to hold them.
| Precision | Weights | Memory needed | Cheapest setup | Per hour | Also fits |
|---|---|---|---|---|---|
| 16-bit | 16.4 GB | 19.7 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 8-bit | 8.2 GB | 9.8 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
| 4-bit | 4.1 GB | 4.9 GB | 1x MI300X (192 GB) Vultr |
$1.85 | 1x H100 $1.99 · 1x MI325X $2.00 |
Memory is the weights at that precision plus 20% for the runtime and a short context; a long context needs more. Prices are the lowest on-demand hourly rates in the SAVRN Index, read Oct 19, 2026.
Keural-Cortex-8B-v1.0-1 on every accelerator the SAVRN Index prices, at every precision
Model Card
By Hika Dalju, published under apache-2.0, revision 3be9e4aeac8e.
Keural Cortex is MKD's local-first Korean-English language-model program for Keural applications. This release is an 8B-parameter, instruction- and preference-aligned checkpoint with a structured tool-calling layer for enterprise workflows. It is designed to run in customer-controlled infrastructure through an OpenAI-compatible vLLM endpoint. The host application—not the model—executes tools, enforces permissions, retrieves live data, and owns user data handling. English. Keural Cortex is a locally deployable AI foundation-model capability built for Korean-English enterprise interaction. It combines efficient 8B-scale inference, controllable OpenAI-compatible serving, and structured…
Read Hika Dalju's full model card
Keural Cortex Foundation Model (8B) — v1.0
Keural Cortex is MKD's local-first Korean-English language-model program for Keural applications. This release is an 8B-parameter, instruction- and preference-aligned checkpoint with a structured tool-calling layer for enterprise workflows.
It is designed to run in customer-controlled infrastructure through an OpenAI-compatible vLLM endpoint. The host application—not the model—executes tools, enforces permissions, retrieves live data, and owns user data handling.
Release status: Public v1.0. Evaluate on your own data, workflows, safety requirements, and target serving configuration before production deployment.
introduction
English. Keural Cortex is a locally deployable AI foundation-model capability built for Korean-English enterprise interaction. It combines efficient 8B-scale inference, controllable OpenAI-compatible serving, and structured tool-call generation so that Keural can connect users to approved enterprise workflows while retaining application-level control of data, permissions, and execution.
한국어. Keural Cortex는 한국어·영어 기반의 엔터프라이즈 AI 상호작용을 위해 설계된 로컬 배포형 파운데이션 모델 역량입니다. 8B 규모의 효율적인 추론, OpenAI 호환 서빙, 구조화된 Tool Call 생성을 결합하여 Keural이 승인된 업무 워크플로와 연동할 수 있도록 지원합니다. 데이터·권한·실행 통제는 모델이 아닌 Keural 애플리케이션 계층에서 관리합니다.
Key strengths
| Strength | What it enables |
|---|---|
| Local-first deployment | On-premise or customer-controlled serving patterns; the application controls networking, credentials, access, and retention. |
| Korean-English interaction | Direct bilingual chat behavior for Keural user experiences. |
| Structured tool calls | Generates OpenAI-compatible calls for search_web, fetch_web_urls, retrieve_documents, and ask_clarification when those tools are provided. |
| Agent-ready architecture | Separates model decision-making from tool execution, allowing the product layer to apply authorization, auditing, retrieval, and real-time data policies. |
| Long-context design | Configuration supports a 65,536-token maximum position length. Serving limits should be set according to available GPU memory and workload. |
| Efficient serving | Tested with vLLM using an OpenAI-compatible /v1/chat/completions interface. |
Model details
| Item | Value |
|---|---|
| Architecture | Decoder-only transformer language model |
| Transformer layers | 36 |
| Hidden size | 4,096 |
| Attention heads / KV heads | 32 / 8 |
| Vocabulary size | 151,936 |
| Maximum position length | 65,536 tokens |
| Weights | BF16 SafeTensors, ~16 GB |
| Alignment path | Continued training, supervised instruction tuning, preference optimization, and Keural agent/tool-contract tuning |
Verified serving smoke tests
The following were verified through vLLM's OpenAI-compatible API in the release environment:
- Plain Korean chat returned a direct answer without emitting a tool call.
- A current-information request generated a structured
search_webcall. - An exhaustive document request generated
retrieve_documentswithexhaustive: true. - An underspecified revision request generated
ask_clarificationwith a question and checklist.
These are smoke tests, not a substitute for domain-specific benchmarking or safety evaluation.
Quick start: vLLM
vllm serve <MODEL_PATH> \
--served-model-name keural-cortex-8b-v1.0 \
--tool-call-parser hermes \
--enable-auto-tool-choice
For text-only chat, send a normal OpenAI Chat Completions request without tools or tool_choice. For agent workflows, provide only the tools your application authorizes, validate all tool arguments server-side, and execute calls outside the model process.
Example tool contract
{
"type": "function",
"function": {
"name": "retrieve_documents",
"description": "Search content in the user's available documents.",
"parameters": {
"type": "object",
"properties": {
"query": {"type": "string"},
"file_ids": {"type": "array", "items": {"type": "string"}},
"exhaustive": {"type": "boolean"}
},
"required": ["query"]
}
}
}
Limitations and responsible use
- Outputs can be inaccurate, incomplete, or inconsistent across prompts and serving configurations. Verify important information.
- The model has no inherent live-web access, private-document access, or authority to take external actions. These capabilities must be supplied and governed by the host application.
- Tool calls are model suggestions. Validate schemas, authorization, user intent, and side effects before execution.
- Do not use this preview as the sole basis for high-stakes medical, legal, financial, safety, employment, credit, or access-control decisions.
Certification, license, and attribution
Certification note. Keural Cortex Foundation Model (8B) v1.0 is based on Qwen3-8B-Base and is released under Apache-2.0. Please retain applicable upstream notices and comply with the license terms.
Citation
@misc{keural-cortex-8b-v1.0,
title={Keural Cortex Foundation Model (8B) — v1.0},
author={MKD AI},
year={2026},
publisher={Hugging Face}
}
Configuration
- Architecture
- Qwen3ForCausalLM
- Context length (tokens)
- 65,536
- Layers
- 36
- Hidden size
- 4,096
- Feed-forward size
- 12,288
- Attention heads
- 32
- Key/value heads
- 8
- Head dimension
- 128
- Vocabulary size
- 151,936
- Model type
- qwen3
Identity and Version
- Repository
- Hikaf143/Keural-Cortex-8B-v1.0-1
- Publisher
- Hika Dalju
- Task
- Text generation
- Modality
- Text
- Library
- transformers
- Parameters
- 8.2B parameters
- Languages
- en, ko
- Revision
- 3be9e4aeac8e80180c8003848c5d945f61548796
- First published
- 2026-10-13
- Last updated
- 2026-10-13
Files and Weights
15 files, 16.4 GB in total. The weights are 1 file totalling 16.4 GB in safetensors.
Weights1 file · 16.4 GB
Configuration2 files · 1.9 KB
Tokenizer2 files · 11.4 MB
Documentation1 file · 6.0 KB
Other8 files · 16.8 MB
Repository1 file · 1.8 KB
Every file
| File | Type | Size | SHA-256 |
|---|---|---|---|
| model.safetensors | Weights | 16.4 GB | ce1121bc9b9e |
| config.json | Configuration | 1.7 KB | — |
| generation_config.json | Configuration | 269 B | — |
| README.md | Documentation | 6.0 KB | — |
| assets/architecture.svg | Other | 2.4 KB | — |
| assets/deployment.svg | Other | 1.8 KB | — |
| assets/keural-cortex-agentic-logo.gif | Other | 12.3 MB | ea6322b85f14 |
| assets/keural-cortex-foundation-cover.png | Other | 1.7 MB | dd8ad6ebf3b4 |
| assets/keural-cortex-hero.svg | Other | 2.6 KB | — |
| assets/keural-cortex-keural-logo.png | Other | 2.7 MB | f766d7885be2 |
| assets/keural-logo.svg | Other | 8.9 KB | — |
| chat_template.jinja | Other | 4.2 KB | — |
| .gitattributes | Repository | 1.8 KB | — |
| tokenizer.json | Tokenizer | 11.4 MB | be75606093db |
| tokenizer_config.json | Tokenizer | 4.7 KB | — |
License and Download
- License
- apache-2.0
- Access
- Open weights, no gate
- Download size
- 16.4 GB
Download from Hika Dalju
Released by Hika Dalju through its official repository on Hugging Face. Read the license.
Memory Requirements
| Precision | Weights in memory |
|---|---|
| As published | 16.4 GB |
| 16-bit | 16.4 GB |
| 8-bit | 8.2 GB |
| 4-bit | 4.1 GB |
Weights only, from the published parameter count; the key-value cache and runtime add to this.
Questions About Keural-Cortex-8B-v1.0-1
How much GPU memory does Keural-Cortex-8B-v1.0-1 need?
About 19.7 GB at 16-bit and 4.9 GB at 4-bit: the weights (8.2B parameters) plus a working margin. A long context needs more.
What is the cheapest GPU to run Keural-Cortex-8B-v1.0-1 on?
At 16-bit, 1x MI300X from $1.85 an hour; at 4-bit, 1x MI300X from $1.85 an hour, at the lowest on-demand prices the SAVRN Index lists.
Can I use Keural-Cortex-8B-v1.0-1 commercially?
Yes. Keural-Cortex-8B-v1.0-1 is released under Apache License 2.0. The Apache License 2.0 is a permissive open-source license. It permits commercial use, modification and redistribution. It requires keeping the license and copyright notices and any NOTICE file, stating significant changes, and it includes an express patent grant from contributors.
What is Keural-Cortex-8B-v1.0-1's context length?
65,536 tokens, from the maximum position embeddings in its published configuration.
Similar Models
Qwen3 is the latest generation of large language models in Qwen series, offering a comprehensive suite of dense and mixture-of-experts (MoE) models. Built upon extensive training, Qwen3 delivers groundbreaking advancements in reasoning, instruction-following, agent capabilities, and multilingual support, with the following key features: - Uniquely support of seamless switching between thinking mode (for complex logical reasoning, math, and coding) and non-thinking mode (for efficient, general-purpose dialogue) within single model, ensuring optimal performance across various scenarios. - Significantly enhancement in its reasoning capabilities, surpassing previous QwQ (in thinking mode) and…
The DeepSeek R1 model has undergone a minor version upgrade, with the current version being DeepSeek-R1-0528. In the latest update, DeepSeek R1 has significantly improved its depth of reasoning and inference capabilities by leveraging increased computational resources and introducing algorithmic optimization mechanisms during post-training. The model has demonstrated outstanding performance across various benchmark evaluations, including mathematics, programming, and general logic. Its overall performance is now approaching that of leading models, such as O3 and Gemini 2.5 Pro. Compared to the previous version, the upgraded model shows significant improvements in handling complex reasoning…
Purpose: Long-memory symbolic reasoning + high-fidelity language generation SymbioticLM-8B is a state-of-the-art hybrid transformer model with built-in symbolic cognition. It combines an 8B Qwen-based transformer with modular symbolic processors and a persistent memory buffer. The model supports both general conversation and deep symbolic tasks such as theorem generation, logical chaining, and structured reasoning with retained memory across turns. - General symbolic reasoning and logical conversation - Code + math proof modeling - Not instruction-tuned (e.g., chat-style inputs may require prompt engineering) - Larger memory buffer may increase CPU load slightly - Symbolic inference is…
English | 简体中文 Qwen3-8B-CC-SFT-v2 is a failure-targeted continued-SFT checkpoint designed to improve coding-agent state preservation and continuation across Claude Code native context compaction. It is initialized from Qwen3-8B-CC-SFT-v1 and trained on compact-aware, context-correct supervision distilled from GLM-5.3 under real Claude Code native compaction. Best observed canonical303 Pass@1: 52.48% (159/303) — 4090 run of this same checkpoint at T=0.5 (temperature selected on an independent dev set; see the table below). Best observed configurations of this same checkpoint: - T=0.5 was selected on an independent 60-task development set (task-id and repo disjoint from canonical303), then…
Ido Levy 1,2 · 1,2 The trained questioner from Asking for What Was Never Requested: Horizontal and Vertical Proactivity in Agents, with its LoRA adapter merged into Qwen3-8B. It is a standard full-weight model: it loads without PEFT and serves with vLLM, SGLang or TGI like any Qwen3-8B. - The adapter, with the results, the training details, the limitations and a complete two-turn - Quantized for llama.cpp, Ollama and LM Studio: This is training seed 1, the adapter at the root of the adapter repository. The merge ran in float32 and the weights are stored in bfloat16. On the adapter card's two-turn example, greedy decoding with this model returns the adapter's output character for character.…