As of November 21, 2026, Level lists 58 live roles at Thinking Machines Lab. The lab released its Inkling 2 Mini model the day before. Every one of those listings posts a US dollar salary range. Most of them are research and engineering roles based in San Francisco.
What happened
On November 20, 2026, Thinking Machines Lab announced Inkling 2 Mini, an open-weights model. The lab says its memory footprint of about 20 GB lets it run on consumer laptops. The post reached the front page of Hacker News the same day.
The lab describes Inkling 2 Mini as a looped transformer. It runs the middle layers of the model three times for every token.
It is a Mixture-of-Experts model with 26.4 billion total parameters. The lab gives 3.4 billion active parameters. Looping raises that to 5.4 billion weight activations per token.
The lab reports that it leads open-weights models under 100 billion parameters on the DeepSWE v1.1 coding benchmark. Its reported result there is 48.2%. The model is available through Tinker, the lab's fine-tuning service, and on Hugging Face.
For a job seeker, the release raises a practical question. Who is this lab hiring, for what work, and at what pay? The Level database gives a measured answer.
What roles is Thinking Machines Lab hiring for?
The answer is mostly engineering and research. Level files 22 of the 58 roles under software engineering. Another 18 roles sit in research.
The remaining 18 roles cover recruiting, product, operations, design, security and sales. You can see the full list on the Thinking Machines Lab company page.
The engineering titles point to the systems behind a model like Inkling 2 Mini. They include training infrastructure, inference, distributed systems, sandboxing and an evaluation platform.
Several titles name Tinker directly. Examples are a Software Engineer, Tinker Infrastructure role and a Research Lead, Tinker, Fine-tuning Science role.
The research titles track the training pipeline. They include pre-training data, mid training, post training, RL scaling, safety and coding agents.
The listings also name the tools the lab expects. PyTorch appears in 22 of the 58 roles. JAX appears in 18 roles.
How much AI is in these jobs?
Most of the work is building AI. Of the 58 roles, 38 are tagged Builds AI, which is 66%. Another 10 roles are tagged Works on AI.
The other 10 roles are tagged Little AI. These are mostly recruiting, people operations and business support roles. No role at the lab sits at Uses AI.
Each label is explained on the levels page. A Builds AI tag means the job itself is about creating or training AI systems.
Fresh demand is a minority of the list. Only 11 of the 58 roles were posted in the last 30 days. The newest is Software Engineer, Tinker Infrastructure, posted on November 17. Training Infrastructure and Frameworks & Libraries engineering roles followed on November 16.
What do Thinking Machines Lab jobs pay?
All 58 listings publish a yearly range in US dollars.
The median low end of those ranges is $300,000. The median high end is $400,000.
Research roles post higher ranges than engineering roles. The 18 research roles have a median range of $350,000 to $475,000.
The 22 software engineering roles have a median range of $300,000 to $400,000. The highest posted ceiling is $500,000. It appears on research and research engineering roles, and on one Engineering Manager role.
Recruiting roles post ranges between $200,000 and $275,000. That matters if you want to join the lab outside research.
These figures describe posted ranges, not actual offers. Equity, which often matters at a young lab, is not part of these numbers.
Can you work remotely?
Sometimes. Of the 58 roles, 16 are marked remote on Level, which is 28%.
Most remote roles are in research and supercomputing. Examples include Research, Safety and Network Engineer, Supercomputing.
The listings still center on one city. Level records San Francisco as the location for 57 of the 58 roles. The remaining role is in New York.
If you need remote work, compare these options with the wider list of remote AI jobs.
What to take from this
First, treat the model launch as a signal of focus, not of new headcount. The release post is a model announcement, not a hiring notice. Level only measures the roles already posted.
Second, the open roles follow the product. The lab builds models, a fine-tuning service and the infrastructure under both. Its 22 engineering and 18 research roles cover that stack.
Third, seniority is listed as mid, senior or lead on 36 of the 58 roles. No role is tagged junior or intern.
For related reading, see Level's posts on open-weight LLM jobs and post-training jobs. You can also browse AI research jobs across all companies.