Trillium Labs pledges to publish AI research as Trillion Labs pushes open models
In brief
- Trillium Labs launched with a stated plan to publish research on AI self-improvement and model behavior.
- Many frontier labs keep their riskiest research private, Crypto Briefing reported.
- Trillion Labs, a separate South Korean company, has built its brand around openness.
- Trillion Labs' Tri series models range from 0.5B to 70B parameters under Apache 2.0.
- Trillion Labs raised a seed round of roughly $4-6 million, according to Crypto Briefing.
What Trillium says it'll publish
The lab's pitch is that it plans to show its work. Trillium says it intends to publish findings on AI self-improvement and model behavior, which Crypto Briefing described as two of the field's most sensitive topics.
That's an unusual stance. According to the same report, many frontier labs keep their riskiest research private, and Trillium is pitching itself as the exception (at least on these two subjects).
Trillion Labs is a different company
The names are close, but they aren't the same lab.
South Korea's Trillion Labs was founded in 2024 by CEO Shin Jae-min, also known as Jay Shin, a former research scientist at Naver. Within one year, it developed and open-sourced a family of language models it calls the Tri series. Those models range from 0.5B to 70B parameters and were released under the Apache 2.0 license.
The 70B model shipped with intermediate training checkpoints. Checkpoints are snapshots of a model taken partway through training, and Trillion Labs describes the release as the first public one of its kind at that scale globally (that's the company's own characterization).
Its Tri-21B model, open-sourced in July 2025, reportedly cut training costs to 1/12 of previous methodologies. Techniques such as XLDA were cited as part of that efficiency push.
NVIDIA ties and seed funding
Trillion Labs presented Korean language model work at NVIDIA GTC 2025. In June 2026, it announced a partnership with NVIDIA to develop industrial "world models" aimed at AI data centers and power infrastructure, which moves the company beyond language models and into infrastructure-focused AI.
Funding is modest by frontier-lab standards. Crypto Briefing reported that Trillion Labs has secured approximately $4-6 million in seed funding from investors including Kakao Ventures and Strong Ventures.
For anyone who's followed web3, the open-weights, open-research approach will look familiar. Trillion's Apache 2.0 licensing and public checkpoints mean outside teams can inspect how a large model develops during training, not just the finished product. Trillium is making a similar promise on research topics that many frontier labs keep private.
Frequently asked questions
Are Trillium Labs and Trillion Labs the same company?
No. Crypto Briefing described South Korea's Trillion Labs as a separate company from Trillium Labs. Trillium is a newly launched lab that plans to publish research on AI self-improvement and model behavior. Trillion Labs was founded in 2024 and has open-sourced its Tri series language models.
What are intermediate training checkpoints?
Checkpoints are snapshots of a model taken partway through training. Trillion Labs released intermediate checkpoints with its 70B model and describes it as the first public release of its kind at that scale globally.
