Skip to content

Measured market guide

JAX machine learning jobs after a DeepSeek V3 rewrite

· 4 min read · Figures from live job ads

Share this guideThreadsLinkedInXEmail

As of November 16, 2026, Level lists 569 live roles that name JAX as a tool. A recent open-source project, a 190-line JAX rewrite of the DeepSeek V3 architecture, drew strong interest on Hacker News. Here is who hires for JAX skills today, and what those roles ask for.

What happened

On November 13, 2026, developer Vaibhaav Tiwari published deepseek-v3-jax on GitHub under an MIT license. The repository rebuilds the model design described in the DeepSeek-V3 technical report. It uses plain JAX, with no Flax, Equinox or PyTorch.

The code covers the main pieces of the paper. It implements Multi-head Latent Attention, a mixture of experts with one shared expert and routed experts, and multi-token prediction. It also reproduces the auxiliary-loss-free load balancing described in the report.

The toy model has about 36 million parameters. The author trained it on a MacBook Air in roughly two and a half hours. In the write-up, the author presents it as an architecture study, not as a strong language model.

The Hacker News thread went up on November 14, 2026. It collected 273 points.

Who hires JAX engineers right now?

Demand is concentrated in a small group of employers. Amazon leads the count with 87 open JAX roles on Level. NVIDIA lists 70 of them. Waymo also lists 70.

Research labs come next. Mistral AI has 23 open JAX roles. Thinking Machines Lab has 18. xAI follows with 13 roles. Anthropic and Cohere list 12 each.

You can browse each employer directly. See the open roles at Amazon, NVIDIA, Waymo and Thinking Machines Lab.

These are not side mentions of a framework. In total, 514 JAX roles sit at the top level, Builds AI. That share is 90%. Most jobs that name JAX involve building or training models. You can read how each level is set on the levels page.

Should you learn JAX or PyTorch?

Learn both. In our data, JAX is rarely asked for alone. Of the JAX roles, 534 also name PyTorch. That is 94% of them.

Only 35 JAX roles on Level do not mention PyTorch at all. So the practical answer is to learn both, with PyTorch first and JAX as a plus.

PyTorch remains the larger market by far. Level lists 2,600+ live roles that name PyTorch. For a wider view of that stack, see our PyTorch jobs page and our earlier piece on CUDA, TPU and PyTorch roles.

Other tools appear often in the same listings. TensorFlow shows up in 260 of the JAX roles. The inference engine vLLM appears in 72 of them.

What do JAX roles pay, and where are they?

Among JAX roles with a posted US dollar range, the median lower bound is $192,000+. The median upper bound is $270,000+.

Those figures come from a subset of listings. Only 296 JAX roles post an upper salary bound in US dollars.

Most roles are in the United States, with 374 listings. That is 66% of the total. The United Kingdom follows with 48 roles. France comes next with 27.

Remote work is possible but not the norm. Level counts 117 JAX roles open to remote candidates. That share is 21%. Our remote AI jobs page tracks the full remote market.

The entry door is real, though narrow. There are 85 JAX role titles that mention an internship. Our AI internships page collects these across every framework. Software engineering is the largest category, with 196 roles. Research comes next, with 167 roles.

Open roles right now

These are live JAX roles on Level, picked by our default ranking. Each one opens the full listing with its level and its source.

569 open roles match, 10 shown. Updated from live job ads.

You can see every role that names the framework on our JAX jobs page. If you are new to JAX, aim first for Builds AI roles that also list PyTorch, since that is where most openings sit.

Method

Source: Level jobs table, read on November 16, 2026, around 10:40 UTC.

Scope: live roles whose tools list includes JAX, across all countries and employers.

Level counts: levels follow our public thresholds, with Builds AI as the highest.

News facts: GitHub repository, author write-up, arXiv report and Hacker News thread, all linked above.

Limits: salary medians cover only listings that post a US dollar range, and tool tags come from listing text.

New AI jobs by email

Want a person to read your CV? Get a free CV review

Browse AI jobs