As of November 5, 2026, Level lists 126 live roles that mention open-weight models, posted by 51 employers. Builds AI roles make up 72 of them, and 45 can be done remotely. Two open-weight releases landed this week: Kimi K3.1 Pro and Ripple-65B Preview.
What happened
Moonshot AI published the weights of Kimi K3.1 Pro on Hugging Face. The model card lists 2.8T total parameters, 104B activated per token and a 1-million-token context window. The news reached the Hacker News front page on November 4.
Ricursive Intelligence also released Ripple-65B Preview. The lab claims the best U.S. open-weight result on the TerminalBench 2.1 coding benchmark, by 0.4 points over Reflection Beam. It also says its small size lets it run on local devices such as NVIDIA DGX Spark. These are the vendors' own claims, not independent tests.
Who is hiring for open-weight work?
Five employers lead the count. Prior Labs has 20 open roles that mention open-weight models. Reflection AI has 9, Harvey has 8, Mistral AI has 8, and Amazon has 7.
Prior Labs alone holds 16% of the 126 roles. Together, those five employers account for 52 of them. The other 46 employers hold the rest.
You can browse the employers directly: Prior Labs, Reflection AI, Harvey, Mistral AI and Amazon. If inference is your angle, our earlier piece on vLLM jobs covers the serving side.
How central is AI to these roles?
Builds AI, Level 4, covers 72 roles, or 57% of the total. Works on AI, Level 3, covers 31 roles, or 25%. Together, 82% of these roles are Level 3 or 4.
Little AI roles number 20, or 16%. Uses AI roles number 3, or 2%. The full definitions are on the levels page.
What skills do the postings ask for?
Tool tags give a concrete picture. PyTorch appears in 37 of the 126 roles. vLLM appears in 21, and Hugging Face appears in 18.
Seniority is the catch. Senior, lead or executive roles number 87, or 69% of the total. Entry-level and intern roles number only 5, and the other 34 roles sit at mid level or have no stated seniority.
With so few junior openings, a public project gives you something concrete to show. Serve an open model with vLLM, measure its latency, and publish the results.
Which job types mention open weights?
Software engineering leads with 32 of the 126 roles. Research follows with 22, sales has 14, and 25 roles fall under the category named other. The remaining 33 roles are spread across smaller categories.
Job titles tell the same story. The word research appears in the title of 25 roles. Titles with inference, serving, infrastructure or platform appear in 21 roles.
Titles that mention machine learning, AI engineer or software engineer appear in 21 roles. Titles with sales, solutions, customer or partner appear in 8 roles. These title searches overlap, so do not add them together.
Read each posting to see whether it covers building, serving or selling open-weight models.
Where are the roles located?
The United States accounts for 70 of the 126 roles, or 56%. The rest are spread across other countries, and the remote filter above applies on top of any location. If you live outside the United States, check the listing page for the country and visa details before you apply.
What do open-weight roles pay?
Only 49 of the 126 roles publish a yearly salary in US dollars. The median midpoint of those ranges is about $251,000.
Fewer than half of the roles show pay, so the median describes a minority of the postings. Check the range on each listing before you plan around it.
Can you work on open weights remotely?
Yes, in 45 of the 126 roles, or 36%. Our remote AI jobs page lists the full set across all topics.
