Vikasit Nano
Smallest general-purpose model. Autocomplete, quick responses, embedded use.
Overview
Vikasit Nano is the smallest model in the lineup — a 0.6B dense transformer for autocomplete, classification, quick Q&A, and on-device / embedded deployment. Runs comfortably on CPU and mobile-class hardware.
Specifications
- Total parameters
- 0.6B
- Architecture
- Dense transformer
- Layers
- 28
- Attention
- GQA (16 query / 8 KV heads), tied embeddings
- Context window
- 32K native
- Vocabulary
- 151,669
- Modalities
- Text in → text out
- License
- Apache 2.0
Capabilities
- Text completion and autocomplete
- Simple classification and extraction
- Runs on CPU / mobile, no GPU required
- Thinking and non-thinking modes
Benchmarks
| Benchmark | Score |
|---|---|
| MMLU-Pro | 24.7 |
| GPQA-Diamond | 27.9 |
| AIME 2025 | 15.1 |
| MATH-500 | 77.6 |
| LiveCodeBench v5 | 12.3 |
| BFCL v3 | 46.4 |
| IFEval | 59.2 |
| HumanEval | N/A |
Instruct numbers from the Qwen3 Technical Report (post-training tables); MMLU-Pro is the base-model figure (instruct tables use MMLU-Redux). Thinking-mode scores shown where modes differ.
Hardware & deployment
| Precision | Memory |
|---|---|
| bf16 | ~1.2 GB |
| INT4 | ~0.4 GB |
Quick start
Vikasit Nano is an open-weight model. Self-host it with any OpenAI-compatible inference server and call it with the OpenAI SDK as shown below.
# pip install openai
import os
from openai import OpenAI
client = OpenAI(
base_url="http://localhost:8000/v1",
api_key="sk-local", # self-hosted servers accept any token
)
resp = client.chat.completions.create(
model="vikasit-nano",
messages=[
{"role": "user", "content": "Explain Vikasit Nano in one sentence."}
],
)
print(resp.choices[0].message.content)Limitations
- Limited multi-step reasoning vs larger models
- Short factual recall; verify critical facts
- Best for narrow, well-scoped tasks
Vikasit Nano FAQ
How much does Vikasit Nano cost?
Vikasit Nano is an open-weight model built on Qwen3-0.6B (Apache 2.0). Self-hosting the weights is free under the Apache 2.0 licence — you pay only for the hardware or cloud GPUs you run it on. Typical deployment fits the memory profiles listed in the hardware section above.
Is Vikasit Nano open weight?
Yes. Vikasit Nano is built on Qwen3-0.6B (Apache 2.0) and distributed under the Apache 2.0 licence, so the weights are openly available for self-hosting, fine-tuning, and commercial use, subject to the upstream licence terms.
How do I run Vikasit Nano?
Because Vikasit Nano is open weight, you self-host it with any OpenAI-compatible inference server (such as vLLM or SGLang) loaded with the Qwen3-0.6B (Apache 2.0) weights, then call it with the OpenAI SDK by setting the base URL to your own endpoint.
What context window does Vikasit Nano support?
Vikasit Nano supports a 32K native context window. It is a 0.6B Dense transformer model — full specifications are listed in the table above.
License & attribution
Apache 2.0
Built on Qwen3-0.6B (Apache 2.0). Upstream copyright, license, and attribution notices are retained.