Arcee AI

Arcee AI is an American artificial intelligence company headquartered in Miami, Florida, founded in June 2023 by Mark McQuade (former Hugging Face and Roboflow), Jacob Solawetz, and Brian Benedict. The company develops the Arcee Foundation Model (AFM) family of small language models for enterprise deployment and maintains mergekit, the open-source model-merging toolkit that has become a standard utility in the post-training community. Arcee AI is reported to be raising a $200 million Series B at a $1 billion post-money valuation as of early 2026, which would mark the company's transition from a Series A small-language-model specialist to a billion-dollar enterprise-AI platform.

At a glance

• Founded: June 2023 in Miami, Florida, by Mark McQuade, Jacob Solawetz, and Brian Benedict.
• Status: Private. Reported $200 million Series B in early 2026 at approximately $1 billion post-money valuation.
• Funding: Cumulative private capital approximately $30 million through Series A, with reported Series B in process. Earlier rounds led by Emergence Capital with participation from Flybridge, Long Journey Ventures, and Wndrco. Strategic round disclosed with Prosperity7 Ventures, Microsoft's M12, Hitachi Ventures, Wipro, JC2 Ventures, Samsung Next, and Guidepoint CEO Albert Sebag.
• CEO: Mark McQuade, Co-Founder and Chief Executive Officer. Former Hugging Face (monetization) and Roboflow (Field Engineering).
• Other notable leadership: Jacob Solawetz (Co-Founder), Brian Benedict (Co-Founder), Charles Goddard (mergekit creator, joined via 2024 acquisition).
• Open weights: Yes. AFM-4.5B and the larger sparse-activated AFM model are released open-weights. The mergekit toolkit is open-source under permissive licensing.
• Flagship outputs: AFM-4.5B (4.5-billion-parameter dense small language model), the data-center-optimized AFM sparse-activated model (approximately 120 to 140 billion parameters with 20 to 30 billion active), mergekit (open-source model-merging toolkit), and the Arcee Cloud enterprise platform.

Origins

Arcee AI was founded in June 2023 by Mark McQuade, Jacob Solawetz, and Brian Benedict, three operators with prior experience at Hugging Face, Roboflow, and other ML-tooling companies. The founding thesis was that enterprise adoption of generative AI was being held back by security, compliance, and deployment concerns that horizontal frontier models from OpenAI and Anthropic could not address by API alone. The company's early product line therefore centered on domain-adaptive small language models that customers could train, host, and operate inside their own perimeter.

The most consequential move of the company's first year was the February 2024 merger with mergekit, the open-source model-merging toolkit created by Charles Goddard. Mergekit had become a widely used utility in the post-training community for combining the weights of multiple fine-tuned models without retraining from scratch, and bringing Goddard and the toolkit in-house gave Arcee a research surface and a name in the open-weights community that few enterprise-focused startups had at the time. The Series A of $24 million followed in mid-2024, led by Emergence Capital with Flybridge and existing seed investors participating.

In 2025 the company shifted from selling fine-tuned variants of third-party open-weights models to training its own foundation models from scratch. The Arcee Foundation Model (AFM) family was the result: AFM-4.5B, a 4.5-billion-parameter dense model targeted at instruction-following workloads on commodity infrastructure, was released first, followed later in the year by a larger sparse-activated mixture-of-experts model designed for higher-throughput data-center deployment. A strategic round in 2025 brought in Prosperity7 Ventures, Microsoft's M12, Hitachi Ventures, Wipro, JC2 Ventures, Samsung Next, and additional investors, anchoring the company's enterprise-distribution thesis through corporate-strategic relationships.

The company is reported to be raising a $200 million Series B at approximately a $1 billion post-money valuation as of early 2026, with the use of proceeds reportedly directed toward training a one-trillion-parameter open-weights foundation model built entirely in the United States.

Mission and strategy

Arcee AI's stated mission is to make enterprise-grade language models that are small enough, controllable enough, and compliant enough for regulated industry deployment, rather than competing on raw capability with frontier models from horizontal labs. The strategic premise is that the long tail of enterprise AI workloads does not need a 1.5-trillion-parameter frontier model and would prefer not to send proprietary data through a third-party API.

The strategy combines three threads. First, the AFM foundation-model family at sizes ranging from 4.5 billion dense parameters up to the sparse-activated data-center model. Second, the open-source mergekit toolkit and adjacent post-training tooling, which positions Arcee inside the practitioner workflow rather than at the API-call layer. Third, Arcee Cloud and direct enterprise relationships through corporate-strategic investors such as Microsoft, Samsung, Hitachi, and Wipro.

The competitive premise reflects Arcee AI's positioning as one of the principal commercial small-language-model labs alongside European peers like Mistral AI and Microsoft's Phi family, with the open-weights mergekit ecosystem providing a distinctive distribution surface relative to closed-API enterprise-AI alternatives.

Models and products

• AFM-4.5B. Dense 4.5-billion-parameter small language model. Open-weights release for instruction-following and enterprise inference workloads on commodity infrastructure.
• AFM sparse-activated model. Larger mixture-of-experts model in the 120 to 140 billion total parameter range with 20 to 30 billion active per token, targeted at data-center enterprise deployments.
• mergekit. Open-source model-merging toolkit. Widely used in the post-training community for combining the weights of fine-tuned models without retraining from scratch.
• Arcee Cloud. Managed platform for enterprises to fine-tune, host, and deploy small language models with controls for data residency and compliance.
• Domain-adaptive language model (DALM) tooling. Earlier-stage product line for fine-tuning open-weights models on customer corpora.

Distribution channels include direct enterprise sales through corporate-strategic partners (Microsoft Azure, Samsung Next portfolio, Hitachi, Wipro), the Hugging Face open-weights distribution surface, and the mergekit open-source community.

Benchmarks and standing

Arcee AI's evaluation framework focuses on small-model efficiency and enterprise-deployment metrics rather than horizontal frontier-model leaderboards. AFM-4.5B has been characterized in technical coverage as a competitive entry in the 4-billion-parameter dense model class against Microsoft Phi-3, Qwen-2.5-3B, and Llama-3.2-3B, with positioning emphasizing instruction-following quality and per-dollar inference economics. The sparse-activated AFM model has been positioned against Mistral Mixtral and DeepSeek MoE peers at the data-center scale.

Industry coverage has consistently grouped Arcee AI with the small-language-model and enterprise-AI segments rather than the frontier foundation-model tier. The open-source mergekit toolkit gives the company a distinctive presence in the open-weights research community that few enterprise-focused startups have replicated.

Leadership

As of May 2026, Arcee AI's senior leadership includes:

• Mark McQuade, Co-Founder and Chief Executive Officer. Former Hugging Face (monetization) and Roboflow (Field Engineering).
• Jacob Solawetz, Co-Founder.
• Brian Benedict, Co-Founder.
• Charles Goddard, mergekit creator and senior research engineer.
• Senior engineering, product, and enterprise-go-to-market leadership building out the Arcee Cloud commercial platform.

Continued senior engineering recruitment has supported the AFM foundation-model program and the reported one-trillion-parameter training plan.

Funding and backers

• Seed (early 2024): $5.5 million, with Wndrco, Long Journey Ventures, and Flybridge participating.
• Series A (mid-2024): $24 million led by Emergence Capital with Flybridge and prior seed investors.
• Strategic round (2025): Disclosed strategic investment with Prosperity7 Ventures, Microsoft's M12, Hitachi Ventures, Wipro, JC2 Ventures, Samsung Next, and Guidepoint CEO Albert Sebag.
• Series B (reported, early 2026): Reported $200 million Series B in process at approximately $1 billion post-money valuation.

Cumulative disclosed private capital approximately $30 million through Series A; reported Series B would push cumulative capital materially higher and establish unicorn status.

Industry position

Arcee AI occupies a distinctive position as one of the principal commercial small-language-model labs with an enterprise-deployment focus, the mergekit open-source toolkit anchoring research-community presence, and a corporate-strategic investor base pointed at large-enterprise distribution. Industry coverage has characterized the company as a leading independent entrant in the enterprise small-language-model segment alongside Mistral AI, Cohere, and AI21 Labs.

The structural risks are two. First, the enterprise small-language-model segment is competitive: Microsoft Phi, Mistral, Cohere, AI21 Labs, and a long tail of fine-tuning specialists all target similar buyers, and the differentiation between AFM-class models from each lab is narrow on most enterprise workloads. Second, the reported one-trillion-parameter training plan would put Arcee in head-to-head competition with frontier-model labs at a capital scale that the disclosed Series B does not yet underwrite.

Competitive landscape

• Mistral AI. Direct enterprise small-and-mid-language-model peer. European headquarters and similar open-weights distribution posture.
• Microsoft Phi family (phi-3, phi-4). Frontier-lab-adjacent small-model program. Microsoft's M12 corporate-strategic relationship makes the dynamic both competitive and partner-aligned.
• Cohere, AI21 Labs. Enterprise-AI peers with longer-running enterprise commercial track records and different model-size positioning.
• Hugging Face. Open-weights distribution-surface alternative; mergekit lives inside the Hugging Face ecosystem and Arcee historically distributes there.
• DeepSeek, Qwen, Kimi. Chinese open-weights small-and-mid-model peers competing on per-dollar inference economics.
• Snorkel AI, Together AI, Fireworks AI. Adjacent enterprise post-training and inference-tooling peers with overlapping go-to-market.

Outlook

• Whether the reported $200 million Series B closes at the $1 billion valuation and the use of proceeds matches the announced training plan.
• The cadence of AFM-family releases through 2026 to 2027 and how the dense-versus-sparse-activated lineup is positioned relative to Mistral and Microsoft Phi.
• Continued evolution of mergekit and adjacent post-training tooling as differentiation in the open-weights research community.
• The enterprise-distribution traction through Microsoft Azure, Samsung Next, Hitachi, and Wipro corporate-strategic relationships.
• The credibility of the reported one-trillion-parameter training plan against the disclosed capital base.

Sources

• Arcee AI official site. Company reference.
• Arcee AI blog: strategic funding round. Strategic round announcement.
• VentureBeat: Arcee AI lands $24M Series A. Series A reference.
• Arcee AI blog: mergekit merger. Mergekit acquisition reference.
• AI CERTs News: Arcee's billion-dollar AI scaling bid. Reported Series B reference.
• mergekit on GitHub. Open-source toolkit.
• Mark McQuade LinkedIn. Co-Founder and CEO reference.