Skip to content
Computer vision

Next 3 Pro

an advanced multimodal model for coding and complex tasks

2 min read

Next 3 Pro is a 27-billion-parameter multimodal model from Lamapi built on the Qwen3.8-27B architecture. It is trained for writing and debugging code, mathematical reasoning, and handling text, images, and video in English and Turkish. The model excels at understanding UI screenshots, diagrams, and technical documentation to help with software design. It is a solid fit for developers and engineers who need a capable local assistant for demanding technical work.

Key features

  • Coding-focused: generates features, writes tests, fixes bugs, and suggests patches.
  • Multimodal input: analyzes images, screenshots, documents, and video clips.
  • Logical reasoning: solves math problems and performs technical analysis under strict constraints.
  • Supports tool calling and agentic workflows.
  • Bilingual optimization: full support for working in both English and Turkish.

What you can do

  • Generate a web app or script from scratch using a text prompt and refine it based on error logs.
  • Analyze a screenshot of a broken UI or diagram and get concrete code fixes.
  • Solve complex math or engineering problems with step-by-step reasoning.
  • Integrate the model into a local development pipeline or a chatbot with multimodal input support.

Example

import os

import torch
import transformers
from transformers import AutoConfig, AutoProcessor

model_id = "thelamapi/Next-3-Pro"
token = os.environ.get("HF_TOKEN")

config = AutoConfig.from_pretrained(model_id, token=token)
model_class = getattr(transformers, config.architectures[0])
model = model_class.from_pretrained(
    model_id,
    token=token,
    dtype=torch.bfloat16,
    device_map="auto",
).eval()

processor = AutoProcessor.from_pretrained(model_id, token=token)
messages = [{
    "role": "user",
    "content": [{
        "type": "text",
        "text": (
            "Write a Python function that sorts a list of dictionaries by "
            "date without mutating the input. Include one small test."
        ),
    }],
}]

inputs = processor.apply_chat_template(
    messages,
    tokenize=True,
    add_generation_prompt=True,
    enable_thinking=False,
    return_dict=True,
    return_tensors="pt",
)

device = model.get_input_embeddings().weight.device
inputs = inputs.to(device)
prompt_length = inputs["input_ids"].shape[-1]

with torch.inference_mode():
    output = model.generate(
        **inputs,
        max_new_tokens=512,
        do_sample=False,
    )

print(processor.tokenizer.decode(
    output[0, prompt_length:], skip_special_tokens=True,
).strip())
Read moreDetails
Base model: Qwen3.8-27B
Size: 27B parameters (merged Transformers checkpoint)
Training: LoRA fine-tune on GSM8K, MBPP Sanitized, and synthetic datasets
GGUF versions and quantization-compatible builds available
Discuss Telegram 1 32

Will it run?

Runs in 4-bit

Full precision does not fit, a quantised version does: quality barely changes down to 8 bits, more noticeably at 4 bits.

  • BF16weights 51 GB · ready files57 GB
  • FP8 / INT8recommendedweights 25 GB29 GB
  • 4-bitweights 14 GB16 GB

Tip: go 8-bit.FP8 / INT8 is within about a percent of full precision in quality, takes half the memory and runs faster on RTX 40/50 and H100. Full precision only pays off for fine-tuning.

Estimate: weights plus working memory. Long videos, high resolutions and long contexts need more.

Run it on Massed Compute

The script downloads the model to the rented machine. How to run it is in the model card.

  1. 1. Rent a machine with a fitting GPU

    Open Massed Compute
  2. 2. Paste the script into the machine's terminal

    Connect over SSH (the Massed Compute panel shows the command) or open a terminal on the machine's desktop.

    HF_TOKEN / CIVITAI_TOKENThe script starts with token lines: paste yours between the quotes. The Hugging Face token is for gated models, the Civitai one for Civitai models.

    # thelamapi/Next-3-Pro
    mkdir -p ~/models
    curl -LsSf https://astral.sh/uv/install.sh | sh
    source "$HOME/.local/bin/env"
    uv tool install "huggingface_hub[cli]"
    hf download thelamapi/Next-3-Pro --local-dir ~/models/Next-3-Pro
    # The model is in ~/models. How to run it is in the model card (link above).
    Add more LoRAs to Next-3-Pro
  3. 3. Use it

    The model is in ~/models. How to run it is in the model card (link above).

Stop the machine in the panel when you are done: it bills by the minute while it runs.

How to speed it up

Picked for this model type. Choose your hardware above to see what fits you exactly.

  • FlashAttention 44.0.0b40from November 18, 2026new release

    fastest exact attention on B200

    Kernels for Blackwell (beta). Exact attention without quantisation — fine for everything, LLMs included.

    How to enable: pip install flash-attn-4; vLLM and SGLang pick it up themselves.

    Earlier generations (2)
    • FlashAttention 3 — The Hopper (H100/H200) version with FP8.
    • FlashAttention 2 — The standard for RTX 30/40/50 and A100: exact attention without the memory cost of long contexts.
  • torch.compile2.15.0from October 28, 2026new release

    10–30% for free

    Compiles the model for your GPU. Slower first run, faster afterwards.

    How to enable: model = torch.compile(model); in ComfyUI the TorchCompileModel node.

Model family

New posts about versions of Next-3-Pro (quants, finetunes, LoRAs) will show up under the bell.

No known relatives of this model yet. All derivatives on Hugging Face

About the model

thelamapi/Next-3-Pro
Author
thelamapi
Category
Computer vision
Released
Posted

On NeyroLearn

Hugging Face

321

Ask AI about this model

Answered by Gemini 4 · the chat is kept only in this tab

Comments

Computer vision

Script2Play-4B

Script2Play-4B