← All companies

Frontier / Company profile

Arcee AI

What Arcee AI does

Arcee AI is a U.S. small-model lab and enterprise AI platform that builds open-weight foundation models and the supporting toolchain required to train, merge, evaluate, and deploy them in controlled environments. The company’s thesis is that enterprises need models they can inspect and own—plus an end-to-end workflow that can operate in private clouds, on-prem, or at the edge—without being forced into vendor lock-in or opaque, closed-model behavior. Arcee’s model family is branded as Trinity and is released under permissive licensing (and, as of May 29, 2026, the company is aligning the Trinity family to OpenMDW-1.1). The company also operates hosted developer and enterprise offerings, including Arcee Cloud (announced alongside its Series A), an API platform, and a “chat for free” experience.

From a product standpoint, Arcee positions Trinity as a standardized model “family” spanning multiple sizes and use cases, while the platform layer provides model orchestration and production tooling for agentic workloads. In particular, the company highlights long-running agent execution support, including NAC (an open-source agent harness for long-running work) released under Apache-2.0. Arcee’s enterprise go-to-market centers on regulated industries and organizations that require deterministic latency, compliance controls, privacy, and deployability on infrastructure they control.

As of the latest official updates, Arcee continues to expand both its foundation-model surface area (e.g., Trinity-Large-Thinking) and its surrounding developer/enterprise ecosystem (e.g., Hugging Face partnership for model/dataset/trace storage; Trinity Builders Program for community inference credits; and a national-science initiative co-developed with the U.S. Department of Energy).

News

Company record

Aug 13, 2026 · Official · Arcee AIIntroducing NAC, an Open-Source Harness for Long-Running Agent Work

Arcee published NAC, an open-source agent harness designed to support long-running sessions by separating temporary execution context from persistent workstream state.

Jul 22, 2026 · Official · Arcee AIGenesis-Science-1: DOE and Arcee announce an open model for scientific research

Arcee announced a joint effort with the U.S. Department of Energy (DOE) called Genesis-Science-1 (GS1), describing it as an American open-weight model for scientific research and detailing Arcee’s and DOE labs’ roles in training, post-training, and evaluation.

Jun 09, 2026 · Official · Arcee AIArcee partners with Hugging Face to make the Hub the exclusive home for models, datasets, and agent traces

Arcee announced a multi-million-dollar strategic partnership with Hugging Face, stating that the Hugging Face Hub will be the exclusive home for everything Arcee builds, including open model releases, private models, datasets, and logged agent traces.

May 29, 2026 · Official · Arcee AITrinity is moving to OpenMDW-1.1

Arcee announced that the Trinity family (including Trinity Nano/Mini, Trinity-Large, and Trinity-Large-Thinking and quantized variants) is moving to OpenMDW-1.1, describing the licensing motivations for distributing diverse model artifacts under a single standard.

Apr 14, 2026 · Official · Arcee AIIntroducing the Trinity Builders Program

Arcee launched the Trinity Builders Program, described as a community credit grant for developers, researchers, and open-source builders working with Trinity models, giving free inference access on the Arcee API.

Apr 01, 2026 · Official · Arcee AITrinity-Large-Thinking: Scaling an Open Source Frontier Agent

Arcee announced the official release of Trinity-Large-Thinking as a frontier open reasoning model for complex long-horizon agents and multi-turn tool calling, stating availability via its API and weights via Hugging Face under Apache-2.0.

Jan 27, 2026 · Official · Arcee AITrinity Large: An Open 400B Sparse MoE Model (architecture + checkpoints)

Arcee released a deep dive on Trinity Large, described as a 400B sparse MoE with 13B active parameters per token and multiple variants (Preview/Base/TrueBase), and discussed the model’s training and sparse routing setup.

Jul 30, 2025 · Official · Arcee AIArcee AI Secures Strategic Investment to Accelerate Enterprise-Grade AI Platform Built on AFM Foundation Models

Arcee announced a strategic funding round led by Prosperity7 Ventures and M12 with participation from Hitachi Ventures, Wipro, JC2 Ventures, Samsung Next, and Guidepoint’s CEO Albert Sebag.

Show 2 earlier updates
Source map · 2 recurring channels · 15 references

Still resolving: Newsroom · Official X

Funding

Latest disclosed valuationNot disclosed
Tracked capital$29.5M3 sourced rounds

What it builds

Trinity (model family)

A standardized open-weight foundation model family that Arcee describes as spanning capabilities from on-device/edge deployments to cloud use, with weights released under permissive licensing.

source ↗
AFM (Arcee Foundation Models) / AFM-4.5B

Arcee’s foundation model family positioned for enterprise deployment needs, including CPU and edge optimization; AFM-4.5B is described as commercially licensed for deployment across public cloud, private infrastructure, and on-premise networks.

source ↗
Arcee Cloud

A cloud version of Arcee’s SLM training/merging/deploying platform, launched as part of the company’s Series A announcement.

source ↗
Arcee API / Arcee Platform (API-based access)

Arcee’s platform for developers to ship via an API or work with models they can inspect and run; the company’s documentation and homepage describe API-based access to model capabilities.

source ↗
Chat for Free (chat.arcee.ai)

A free chat experience linked from Arcee’s main product navigation, intended to let users interact with the company’s offerings through a hosted UI.

source ↗
Trinity Builders Program

A community program providing developers and researchers free inference access credits for Trinity models on Arcee’s API, with the company stating that the program uses the same secure API infrastructure as commercial tiers and that program data is not retained or used for model training.

source ↗
NAC (open-source agent harness)

An open-source runtime/harness for long-running agent work, designed to separate temporary action context from persistent state and to support long-horizon multi-hour or multi-day sessions.

source ↗
Open Source Catalog (open-weight models and toolkits)

Arcee’s catalog page for open-source/open-weight offerings, including references to the company’s open tooling and model ecosystem components.

source ↗

Milestones & partnerships

Aug 13, 2026
Released NAC (open-source long-running agent harness)

Arcee published NAC as an open-source runtime intended to support longer-horizon agent work by separating action context from persistent state.

Jul 22, 2026
Genesis-Science-1 announced with the U.S. Department of Energy

Arcee and DOE announced GS1 as an open-weight model for scientific research with defined roles for compute/training, workbenches, and evaluation.

U.S. Department of Energy
Jun 09, 2026
Hugging Face partnership for exclusive storage and distribution of Arcee artifacts

Arcee stated that the Hugging Face Hub becomes the exclusive home for models, datasets, and agent traces produced by the company.

Hugging Face
May 29, 2026
Trinity licensing alignment to OpenMDW-1.1

Arcee announced a move of the Trinity family (including quantized variants) to OpenMDW-1.1.

Linux Foundation
Apr 14, 2026
Launched Trinity Builders Program

Arcee launched a community credit grant for inference access to Trinity models via Arcee’s API, including program terms about secure infrastructure and non-retention for training use.

Apr 01, 2026
Official release of Trinity-Large-Thinking

Arcee announced Trinity-Large-Thinking as an open Apache-2.0 frontier reasoning model for long-horizon agents with multi-turn tool calling, distributed via its API and weights on Hugging Face.

Hugging Face
Jul 17, 2024
Series A closes; Arcee Cloud launches

Arcee announced its $24M Series A, the appointment of Julien Simon, and the release of Arcee Cloud (hosted SaaS).

Emergence Capital