As of November 15, 2026, Level lists 13 live jobs that mention quantization, and every one is rated Builds AI (Level 4). Nine employers are hiring, led by NVIDIA with four roles. Six roles publish a US dollar salary range, with a median midpoint of $199,350. Here is what the work involves and who hires.
What happened
On November 11, Unsloth updated its Dynamic quantization (V3.1, preview) for Qwen3.8-27B GGUF files. The Unsloth documentation says it improves accuracy by 5 to 10 percent at the same size on most benchmarks, including Terminal Bench and DeepSWE.
The same page says the 27B model runs locally on 17GB of RAM or VRAM. Quantization is the technique that makes this possible. It stores a model's weights at lower precision, so the model is smaller and cheaper to serve.
For a job seeker, the question is simple. Who pays people to do this work, and what do they ask for?
How many quantization jobs are open right now?
As of November 15, 2026, Level lists 13 live jobs that mention quantization. That is a narrow niche. Level lists 505 live roles that mention inference, so quantization covers about 3% of that volume.
All 13 roles are Builds AI (Level 4). That means the model or its serving stack is the product. See how the levels are defined. You can browse every role at Level 4.
Only 3 of the 13 roles are remote, or 23%. The rest are tied to an office. Seven are in the United States, two in China, two in India, and one each in Israel and France. For more remote options, see remote AI jobs.
Who is hiring for quantization work?
Nine employers share the 13 roles. NVIDIA has four, in India, Israel and China.
Nuro has two, both in ML infrastructure and optimization in Mountain View. The other employers have one each: Runware, Netskope, Thales, SambaNova Systems, OpenAI, Thomson Reuters and Ultralytics.
The mix is wide. Chip and systems companies sit next to an autonomous driving company, a security vendor and a computer vision startup.
Some concrete examples: NVIDIA is hiring a Senior Performance Engineer for LLM inference frameworks in Israel. Nuro lists an ML Infrastructure, Optimization engineer in Mountain View, at $160,360 to $240,540. Runware lists a remote Principal Engineer for model performance in the United States.
See the NVIDIA company page and the Nuro company page for their other open roles.
What skills do these jobs ask for?
The tool tags on the 13 listings point to one stack. PyTorch appears in 12 of them. vLLM appears in 7.
Six listings tag a TensorRT variant, such as TensorRT-LLM. JAX appears in 3 listings, and llama.cpp and SGLang in 2 each. Hugging Face and ONNX appear once.
Only 1 of the 13 titles contains the word, and that one is in French (quantification, at Thales). The others say model performance, inference optimization, TPU compilers or system software. Searching the title alone misses most of these roles.
For related skills, read our guides to inference engineer jobs and vLLM jobs.
What do quantization jobs pay, and can beginners apply?
Six of the 13 roles publish a yearly salary range, all in US dollars. The ranges run from $110,000 to $500,000. The median midpoint is $199,350.
These are midpoints of advertised ranges, not guaranteed offers. The highest range is a senior OpenAI role on TPU compilers, at $380,000 to $500,000.
Beginners have a few doors. Three of the 13 roles are open to early-career candidates. They are an NVIDIA internship in Shanghai, a Thales internship in France and a SambaNova new graduate role in the United States.
Eight of the 13 listings are senior, one is lead and one is mid level. The salary ranges tell the same story, with the SambaNova new graduate role at $131,200 to $196,800. The market leans senior, and 7 of the 13 listings tag vLLM, the open source serving engine. See AI internships for more entry points.
