Arcee AI releases Trinity-Large-Thinking, a 399B-parameter text-only reasoning model under an Apache 2.0 license, allowing full customization and commercial use
The baton of open source AI models has been passed on between several companies over the years since ChatGPT debuted in late 2022 …
VentureBeatCarl Franzen
Context & Ripple Effects
Arcee had already established the Trinity line with a roughly 400B-parameter open-weight release in January; this reasoning-focused version turns that earlier model family into a more explicitly commercializable offering through Apache 2.0 terms. The move also aligns with the company’s stated ambition to train a substantially larger open-weight model, reported while it was seeking new financing.
The story matters because licensing, not parameter count alone, determines whether enterprises and developers can adapt a model into proprietary products. It follows earlier large-model releases such as AI2's 405B-parameter Tulu 3 while making permissive reuse central to Arcee’s positioning.
First-order effects
Developers and enterprises can modify, deploy and commercialize Trinity-Large-Thinking without a proprietary-model access agreement, making Arcee’s model immediately usable as a customizable text reasoning base.
Arcee gains a clearer route to distribution through downstream builders and hosting ecosystems, extending the earlier Trinity Large open-weight release into a reasoning-oriented product line.
Second-order effects
Open-model vendors competing for enterprise adoption face greater pressure to pair model capability with permissive terms, clear deployment rights and customization options rather than relying solely on benchmark claims.
Buyers evaluating text reasoning systems gain another self-hostable option, strengthening procurement leverage against closed API offerings where control, deployment flexibility or commercial reuse matters.
Third-order effects
If large, permissively licensed models continue to improve, value may shift from exclusive access to base models toward fine-tuning, inference infrastructure, integration and domain-specific data.
The pattern points to a more bifurcated model market: proprietary providers can retain customers seeking managed services, while open-weight suppliers compete to become the adaptable foundation layer—though actual adoption will depend on performance and operating costs.
The trend: This is one data point in the industrialization of open-weight AI, where permissive licenses turn frontier-scale models into inputs for commercial software stacks.
Today we're releasing Trinity-Large-Thinking. Available now on the Arcee API, with open weights on Hugging Face under Apache 2.0. We built it for developers and enterprises that want models they can inspect, post-train, host, distill, and own. [video]
the best American open-source model ever just dropped, and it costs less than $1 per million tokens i feel like more people should be talking about this
Today we drop Trinity-Large-Thinking. SOTA on Tau2-Airline, frontier-class on Tau2-Telecom, and the #2 model on PinchBench, right behind Opus. On BCFLv4, we're in the mix with the best. 26 people with under $50M raised and a ruthless pursuit of greatness. What this team just
We're excited to support @Arcee_ai's Trinity-Large-Thinking — a frontier open reasoning model Purpose-built for the agents people are actually running in production. Proud to have supported with our infra and post-training stack including prime-rl and verifiers.
This is a noteworthy release. I don't think there has been been a real open source model from the US that is this close to the frontier, ever. Looking forward to trying it out.
Preview showed us where the demand was going. People were already running Trinity-Large-Preview in real agent workflows, with long-horizon tool use and production constraints. So over the last two months, we pushed our SFT and RL stack to meet that moment. [image]
Trinity-Large-Thinking by @arcee_ai has been added to Design Arena! The current leading open model is GLM 5 by @Zai_org. Huge congrats to the @arcee_ai team for this release! [image]
More important than any one score, Trinity-Large-Thinking is a major step up from Preview in the places that matter most for agents: better multi-turn tool use better context coherence cleaner instruction following more stable long-running behavior
Two things I'm particularly proud of here: 1. The pretraining data are derived entirely from publicly-available tokens. 2. No closed-source models were used in any part of the pretraining data curation pipeline.
Our focus was clear. Build a model that stays coherent across turns, uses tools cleanly, follows instructions under constraint, and is efficient enough to serve at scale. That is the bet behind Trinity-Large-Thinking.
Trinity-Large-Thinking running on M3 Ultra 512GB in 4bit using MLX! 🚀 Text Generation at 48 toks/s Peak Mem 224GB Quantizations (4bit and 5bit) upload in progress on HF mlx-community! 🚀 [video]