GLM-5.2 is live. #1 throughput on OpenRouter, pay-per-token on FriendliAI. Try it today ➜
vectionlabs
Available on FriendliAI
Run this model inference on single tenant GPU with unmatched speed and reliability at scale.
Model Details
Model Provider
vectionlabs
Model Tree
Base
this model
Input Modalities
Text
Image
Video
Output Modalities
Text
Supported Functionality
Dedicated Endpoints
Ships patches, not suggestions. Reads before it writes. Finds the cause instead of the symptom, then changes the smallest thing that fixes it.
Lives in a terminal. Plans the sequence, checks the result of each step before taking the next, and recovers from a failure rather than repeating it.
Sees what you see. Screenshots, diagrams, mockups, stack traces in an image — first-class input, not an afterthought.
Thinks without being told to. Reasoning is native and on by default. You never have to write think step by step — and you should not: this model reasons on its own, and asking makes it perform instead.
python
from transformers import AutoModelForImageTextToText, AutoTokenizerrepo = "vectionlabs/Salience-1.5-Nano"tok = AutoTokenizer.from_pretrained(repo)model = AutoModelForImageTextToText.from_pretrained(repo, dtype="auto", device_map="auto")messages = [{"role": "user", "content": "Find and fix the off-by-one in this loop: ..."}]text = tok.apply_chat_template(messages, tokenize=False, add_generation_prompt=True)inputs = tok(text, return_tensors="pt").to(model.device)out = model.generate(**inputs, max_new_tokens=2048)print(tok.decode(out[0, inputs.input_ids.shape[1]:], skip_special_tokens=True))
~19 GB in fp16, ~6.5 GB quantized. Pass enable_thinking=False when you want the answer and
not the reasoning.
| Parameters | 9.65B dense |
| Modalities | text, image, video to text |
| Context | 262,144 tokens |
| Precision | bfloat16 |
| License | Apache-2.0 |
Nano is small: it will lose to frontier-scale models on the hardest problems, and it can be confidently wrong — verify what matters. Review generated code before running it. Built on Qwen3.5 (Apache-2.0) with work from Ornith-1.0-9B (MIT) and tmax-9b (Apache-2.0, used per Ai2 Responsible Use Guidelines).