What Arcee.ai does

Arcee AI is a privately held, U.S.-based open-weight AI lab founded in 2023 by former Hugging Face engineers, headquartered in San Francisco with a registered entity in Venice, Florida, and roughly 30 employees (14 in research). The company develops the Trinity family of small language models — Nano (6B total / 1B active), Mini (26B / 3B active), and Large (400B / 13B active), plus a Large-Thinking reasoning variant — built on Arcee's proprietary AFMoE sparse mixture-of-experts architecture with sigmoid routing and gated attention. Trinity-Large-Thinking, the largest open-weight MoE model released by a U.S. company at the time, was trained for approximately $20M over 33 days on 2,048 Nvidia B300 GPUs using 8T+ tokens of training data curated with DatologyAI, and posts agentic-reasoning scores of 88 on Tau2-Airline, 91.9 on PinchBench, and 96.3 on AIME25.

The product portfolio extends beyond base models. Arcee Conductor is an intelligent router that dynamically selects the optimal model per prompt from a catalog of Arcee SLMs and third-party frontier models to cut cost per prompt by up to 99%. Arcee Orchestra is an agentic workflow platform for building multi-step agents on top of the model family. Arcee also maintains three open-source toolkits — MergeKit (with the proprietary Arcee Fusion method), DistillKit, and EvolKit — that drive developer adoption and ecosystem tooling, and IBM Research publicly used MergeKit during the development of Granite 4.0.

Arcee monetizes through a hybrid mix: token-based pay-as-you-go API consumption (OpenAI-compatible, with prepurchased tokens), enterprise model build/deployment/integration services via the "Work With Us" intake, redistribution through third-party Model Providers (OpenRouter, Together.AI, Clarifai, Hugging Face) under a Model Provider Agreement, and a freemium open-source funnel anchored by the free Playground and the Trinity Builders Program credit grant. Customers span Fortune 500 enterprises in financial services, insurance, retail, and education (with disclosed outcomes of 96% deployment-cost reduction at a Fortune 500 financial-services firm and 82% cost reduction plus 63% performance uplift at a global insurer), U.S. government and regulated buyers seeking sovereign alternatives to Chinese open-weight models, and a long tail of self-serve developers on Hugging Face and OpenRouter. The company has raised approximately $50M cumulatively across a seed round, a $24M Series A led by Emergence Capital (July 2024), and a strategic round in July 2025 (M12, Prosperity7 Ventures, Hitachi Ventures, JC2, Wipro, Samsung, Guidepoint), and was reported in February 2026 to be negotiating a round exceeding $200M at a $1B+ valuation.

Name

Arcee.ai

Legal name

Arcee AI, Inc.

Website

https://arcee.ai

Company type

Private

Founded year

2023

Operating status

Operating

Headcount range

11–50 employees

Short description

Arcee AI is a privately held, San Francisco-based open-weight AI lab founded in 2023 that develops the Trinity family of small language models (Nano, Mini, Large, Large-Thinking) on a proprietary sparse MoE architecture, serving Fortune 500 enterprises, government and regulated buyers, and self-serve developers via API, open-source downloads, and partner redistribution.

Ownership category

akta.pro rank

Product category

Open-weight Foundation Models

NAICS

Software Publishers (5132), Software Publishers (513210), Custom Computer Programming Services (541511), Computing Infrastructure Providers, Data Processing, Web Hosting, and Related Services (51821)

SIC

Services-Prepackaged Software (7372), Services-Computer Programming, Data Processing, Etc. (7370)

akta.pro primary industry

Open-Source Model Ecosystems & Model Marketplaces (HDAAACAM)

akta.pro secondary industries

Model Serving, Inference & Deployment Platforms (APIs, Edge/On-Prem) (HDAEANAC), AI Integration & Orchestration Platforms (Connectors, Workflow, iPaaS for AI) (HDAEANAI), AI Application Enablement Platforms (Copilot/Agent Frameworks, SDKs) (HDAEANAJ), Agents & Autonomous Workflows (Tool Use, Planning, Multi-Agent) (HDAAACAF)

Keywords

• Small language models
• Open-weight AI
• Enterprise AI platform
• AI model deployment
• Foundation model API

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Offices3 records

Markets served

GTM type

B2B

Offering type

Software

Cost components

Technology or R&D, Personnel, Infrastructure, Marketing or Sales

Revenue model

1. Token-based API consumption: Customers prepay for tokens through their Arcee Account and consume them in real time via the OpenAI-compatible API; tokens are non-refundable, non-transferable, and not redeemable for cash. Pricing denominated in U.S. dollars and subject to change.
2. Enterprise model build, deployment, and integration services: Custom-built or customized models, production deployment, and integration support for enterprise customers via the 'Work With Us' intake, including dedicated support.
3. Model Provider redistribution: Third-party Model Providers (e.g., OpenRouter, Together.AI, Clarifai, Hugging Face) resell access to Arcee models under a Model Provider Agreement, with Arcee retaining underlying rights.
4. Open-source community / freemium funnel: Free open-weight downloads under Apache 2.0 / OpenMDW-1.1 and a free Playground serve as a top-of-funnel land motion that converts developers into paying API and enterprise customers.

Pricing tiers

Go-to-market motion5 records

Distribution channels5 records

Marketing channels10 records

Core offering

Arcee.ai builds and releases open-weight small and large language models under the Trinity family (Nano, Mini, Large, Large-Thinking) and AFM-4.5B, all built on its proprietary AFMoE sparse Mixture-of-Experts architecture. It monetizes these models through an OpenAI-compatible hosted API (token-based pay-as-you-go consumption), an intelligent model router called Arcee Conductor, an agentic workflow platform called Arcee Orchestra, and enterprise custom-build, deployment, and integration services. Open-source toolkits (MergeKit, DistillKit, EvolKit) and Hugging Face downloads serve as a freemium funnel into the paid API and enterprise offerings.

Differentiator

Problem solved

Functional benefit

Brands

• Trinity: Arcee's flagship family of open-weight foundation models (Trinity Nano, Trinity Mini, Trinity Large Preview, Trinity Large Thinking) released under Apache 2.0, providing standardized capabilities from edge to cloud.
• Arcee Foundation Models (AFM)
• Arcee Conductor
• Arcee Orchestra
• MergeKit
• DistillKit
• EvolKit

Products and services

• Arcee Conductor Intelligent model router that routes each prompt to the optimal model from a catalog including Arcee models, third-party frontier models, and cloud LLMs, cutting cost per prompt by up to 99% without sacrificing quality. Sold to enterprise developers and platform teams building AI applications.
• Arcee Orchestra End-to-end SLM-powered agentic AI platform for building agentic workflows with tool use, function calling, structured outputs, and multi-step orchestration. Targeted at enterprise teams deploying AI agents.
• Arcee Enterprise Build & Deployment Services Professional services offering covering custom model builds, customization, production deployment, and integration for enterprise customers, acquired through a 'Work With Us' contact form and delivered under multi-year contracts with dedicated support.
• Arcee Hosted API (Trinity & AFM Models) OpenAI-compatible hosted API providing pay-as-you-go token consumption access to Arcee's Trinity family (Nano, Mini, Large, Large-Thinking) and AFM-4.5B models. Customers prepay for tokens through their Arcee Account and consume them in real time; tokens are non-refundable, non-transferable, and not redeemable for cash.
• Arcee MergeKit Open-source model merging toolkit supporting arbitrary transformers and raw PyTorch, including the proprietary Arcee Fusion method that scores parameter importance via divergence and selectively integrates meaningful changes. Used by IBM Research during Granite 4.0 development.
• Arcee DistillKit

Quantifiable outcome

• 96% reduction in deployment costs for a Fortune 500 financial services customer, with 23% improvement in benchmark rankings
• +5 more outcomes

Named customers5 records

Segments4 records

Ideal customer profiles4 records

Technology focussed Yes

API detail

Has API

No

API docs

API detail

Core technology

AI maturity

App detail

Feature9 records

Partnerships

21 partnerships are on record, tiered flagship / exclusive distribution and storage partnership, major oem/edge integration partner, primary launch and hosting partner for the trinity family, strategic workflow and agentic partner, adjacent product partnership, strategic hyperscaler partnership (strategic collaboration agreement), core compute partner, core silicon partner for edge/on-prem, core silicon partner for on-device inference, major managed inference partner, major managed inference partner and benchmark signal, cloud distribution partner, strategic ecosystem partner, core data partner, core u.s. compute partner, adjacent marketing partnership, adjacent technology partner for retail demo, developer tooling integration partner and tooling collaboration (open-source mergekit).

• Hugging Faceflagship / exclusive distribution and storage partnershipStrategic or Co-development Partner · 9 June 2026Multi-million-dollar strategic commercial partnership making Hugging Face the exclusive storage and distribution platform for all of Arcee's models, datasets, and agent traces (public and private). Arcee becomes the first major American AI lab to replace AWS S3 with Hugging Face Buckets. The arcee-ai organization is a flagship on the Hub with planned co-developed releases and deeper integrations throughout the year.
• Ciscomajor oem/edge integration partnerTechnology or Integration · 19 February 2026Co-deployed Arcee small language models on Cisco Unified Edge hardware for retail store-floor AI at the NRF Big Show, in combination with Intel CPUs and Proto Hologram's interface, to deliver real-time, low-latency conversational assistance.
• Clarifaiprimary launch and hosting partner for the trinity familyChannel Partner/ Reseller/ Distributor · 5 December 2025Clarifai is the primary launch partner and hosting provider for the Trinity family (Trinity Nano 6B, Trinity Mini 26B, and the larger Trinity Large in development). Models are served via Clarifai's Compute Orchestration platform through an OpenAI-compatible API with autoscaling, GPU fractioning, and cost-efficient inference.
• Zervestrategic workflow and agentic partnerGTM or Marketing Partner · 27 June 2025Zerve integrates Arcee Conductor to automate AI model selection, and the two companies co-market agentic data science workflows combining Zerve's data science canvas with Arcee Conductor.
• AngelQadjacent product partnershipStrategic or Co-development Partner · 13 February 2025Joint launch of 'KidRails for LLMs', a safety/guardrail framework built on Arcee AI language models for child-facing applications.
• Amazon Web Services (AWS)strategic hyperscaler partnership (strategic collaboration agreement)Strategic or Co-development Partner · 25 November 2024Strategic Collaboration Agreement to deliver advanced, tailored small language models (SLMs) to the market through AWS infrastructure. Uses Amazon SageMaker HyperPod and AWS Graviton for training and inference; AWS EC2 Capacity Blocks for ML used to reserve GPU clusters for training.
• NVIDIAcore compute partnerTechnology or IntegrationNVIDIA GPUs (B300, H100, H200, A100, L40S, Blackwell) power Arcee model training and inference, including the 2,048-B300 cluster used for Trinity Large and the AWS EC2 Capacity Blocks program.
• Intelcore silicon partner for edge/on-premTechnology or IntegrationJoint work on optimizing Arcee Foundation Models on Intel CPUs, with retail edge AI demo at NRF combining Arcee SLM, Intel CPUs, Cisco Unified Edge, and Proto Hologram.
• Armcore silicon partner for on-device inferenceTechnology or IntegrationCo-published exploration of small language model inference on Arm CPUs, targeting on-device and edge AI workloads.
• Together.AImajor managed inference partnerChannel Partner/ Reseller/ DistributorArcee SLMs (including AFM-4.5B) are available on Together.AI via the Together Playground, the Together Python SDK, and an OpenAI-compatible API endpoint.
• OpenRoutermajor managed inference partner and benchmark signalChannel Partner/ Reseller/ DistributorTrinity-Large-Preview, Trinity Mini, and Trinity-Large-Thinking are served on OpenRouter; early OpenRouter usage ranked Trinity among the most-used open models in the U.S.
• Digital Oceancloud distribution partnerChannel Partner/ Reseller/ DistributorDigital Ocean listed as a partner for Arcee models and infrastructure.
• Microsoftstrategic ecosystem partnerTechnology or IntegrationMicrosoft listed as a partner on the Arcee home page; Microsoft Clarity is used for session analytics on the Arcee site.
• DatologyAIcore data partnerStrategic or Co-development PartnerStrategic data partnership: DatologyAI curates Arcee's training corpora, including 8T+ tokens for AFM-4.5B and 10T tokens for Trinity Mini and Trinity Nano, using model-based quality filtering, embedding-based curation, target distribution-matching, source mixing, and synthetic data.
• Prime Intellectcore u.s. compute partnerTechnology or IntegrationPrime Intellect provides U.S.-based compute for Arcee training runs, including the 512 H200 GPU cluster used for Trinity Nano with HSDP parallelism.
• RWS (Tridion)adjacent marketing partnershipGTM or Marketing PartnerJoint blog on building smarter AI workflows by combining Arcee AI models with RWS Tridion.
• Proto Hologramadjacent technology partner for retail demoTechnology or IntegrationCo-demonstrated a holographic store-floor AI assistant (Kaleigh) at the NRF Big Show, combining Arcee AI's small language model, Cisco Unified Edge, Intel CPUs, and Proto Hologram's interface.
• Kilo Codedeveloper tooling integration partnerTechnology or IntegrationListed as a partner on the Arcee home page; appears to be an AI coding tool integrated with Arcee models.
• Clinedeveloper tooling integration partnerTechnology or IntegrationListed as a partner on the Arcee home page; appears to be an AI coding agent integrated with Arcee models.
• Open Codedeveloper tooling integration partnerTechnology or IntegrationListed as a partner on the Arcee home page; appears to be an open-source AI coding tool integrated with Arcee models.
• IBM Researchtooling collaboration (open-source mergekit)Technology or IntegrationIBM Research used Arcee MergeKit during development of Granite 4.0, leveraging the tool to run ablation experiments and select merged checkpoints for the enterprise-focused open-source model.

Scale indicators15 records

Recent moves6 records

Expansion highlights7 records

Broad incumbents

• OpenAI: Closed-source frontier model provider and dominant enterprise AI incumbent; comparable as the primary cost and capability benchmark Arcee's cost-reduction claims are measured against in enterprise pitches.
• Meta (Llama): Meta's Llama family is the dominant open-weight LLM benchmark; comparable as the largest U.S. open-weight competitor with vastly greater compute resources and headcount that sets the capability bar Arcee's Trinity family must meet.
• Alibaba (Qwen): Alibaba's Qwen family is a leading Chinese open-weight model line spanning edge to cloud sizes; comparable to Arcee on open-weight distribution scale, MoE architectures, and as another sovereign-AI counterpart.
• Anthropic: Closed-source frontier model provider with strong enterprise traction via Claude; comparable as the primary frontier-quality benchmark Arcee's Trinity family competes with in enterprise agent and reasoning workloads.

Others

• Together.AI: Managed inference partner that resells Arcee SLMs and also develops and serves its own open-weight models; comparable as an open-model ecosystem participant with overlapping inference and developer-tooling capabilities.
• Hugging Face: Platform partner that distributes Arcee's models exclusively on its Hub; comparable as both a strategic ally and a potential competitor because Hugging Face also publishes its own models and offers inference infrastructure.

Direct peers

• DeepSeek: Chinese open-weight model lab known for cost-efficient MoE training at frontier capability; directly comparable to Arcee as an open-weight rival whose existence underpins Arcee's sovereign-AI positioning.
• Mistral AI: European open-weight foundation model lab releasing competitive LLMs (Mistral, Mixtral MoE) with enterprise and developer go-to-market; directly comparable to Arcee on open-weight model strategy, MoE architectures, and enterprise SLM positioning.
• IBM (Granite): IBM's Granite family of enterprise open-weight models (which used Arcee MergeKit during Granite 4.0 development) is directly comparable on open-weight enterprise distribution and sovereign-AI positioning for regulated buyers.
• Cohere: Enterprise-focused AI lab offering customizable language models with deployment flexibility; directly comparable to Arcee on enterprise SLM positioning, sovereign-deployment messaging, and developer/API GTM.

Market position

Strengths2 records

Weaknesses5 records

Competitive moat7 records

Key risks6 records

Key highlights7 records

Customer concentration

• GitHubgithub.com/arcee-ai
• Xx.com/arcee_ai
• LinkedInlinkedin.com/company/arcee-ai

Revenue estimate

Valuation estimate

Number of profiles

Profiles5 records

Funding overview

Funding rounds5 records

Investors20 records

Funding detail is available on the Subscription and Enterprise plan.Contact sales →

M&A

Investments

M&A and investment is available on the Subscription and Enterprise plan.Contact sales →

Frequently asked questions about Arcee.ai