Cohere

It is remote-first and open globally, with no relocation required, and it explicitly does not require a record of published papers – the program states it seeks candidates with strong engineering skills and creative thinking, including those with little or no publication experience (Research Scholar listing). Scholars are matched to a research project only after selection, so it works as a paid entry into frontier ML research rather than a fixed-curriculum internship – worth prioritising for someone targeting a research career without a conventional publication track record.

Cohere builds enterprise foundation models rather than a consumer chatbot, and several specifics set it apart for early-career applicants:

• it was founded in 2019 in Toronto by former Google Brain researchers and keeps a Canadian base alongside San Francisco, unusual among frontier labs
• it releases open-weight models: Command A+ shipped under a permissive Apache 2.0 license, aimed at enterprise and public-sector "sovereign" deployment (release)
• its Aya family targets multilingual coverage, with recent open-weight models supporting 70+ languages (Aya release)
• it runs a dedicated research arm, Cohere Labs, with the Scholars Program and a Catalyst grant program as explicit on-ramps for outside researchers

For an early-career researcher, open-weight publishing plus a formal scholars-and-grants pathway means the work is visible outside the company and there is a defined way in that does not hinge on prior industry experience.