Salesforce AI Research

Salesforce AI Research is the artificial intelligence research division of Salesforce, the publicly listed American enterprise customer-relationship-management (CRM) software company headquartered in San Francisco. The research division is led by Silvio Savarese, Chief Scientist of Salesforce AI Research and concurrent Stanford computer-science professor. Salesforce AI Research has developed a portfolio of open-source and closed-weights AI models including the CodeT5 and CodeGen code-generation models, the BLIP and BLIP-2 multimodal vision-language models, the XGen and xGen-Sales language models, the xLAM Large Action Models for agentic AI, the SFR-Embedding embedding models, and the Moirai time-series foundation model. The research division provides the foundational AI capability underlying Salesforce's Einstein AI platform and the Agentforce agentic AI platform.

At a glance

• Founded: Salesforce was founded in 1999 by Marc Benioff, Parker Harris, Dave Moellenhoff, and Frank Dominguez. Salesforce AI Research as a dedicated research division was established in subsequent decades; the modern research organization scaled substantially after the 2016 acquisition of MetaMind (Richard Socher's AI startup).
• Status: Subsidiary research division of Salesforce. Salesforce is publicly listed on the New York Stock Exchange (CRM) and is one of the largest enterprise software companies globally.
• Funding: Salesforce internal R&D budget. Salesforce's annual research-and-development investment exceeds $5 billion.
• CEO: Silvio Savarese, Chief Scientist of Salesforce AI Research. Concurrent appointment as Stanford computer-science professor. Marc Benioff provides senior corporate strategic leadership as Salesforce Chair and CEO.
• Other notable leadership: Senior research-program leadership across Salesforce AI Research's principal program areas including foundation models, code generation, multimodal AI, embeddings, agentic AI, and time-series AI. Earlier senior leadership included Richard Socher (Chief Scientist 2016 to 2020; left to found You.com).
• Open weights: Yes, partial. CodeT5, CodeGen, XGen, xLAM-1B, SFR-Embedding, and other research-stage models have been released open-weights through Hugging Face under the Salesforce organization. Production-grade models powering Einstein and Agentforce are closed-weights.
• Flagship outputs: xLAM (Large Action Models for agentic AI; xLAM 8B, xLAM-1B), xGen-Sales (proprietary sales-task model powering Agentforce), BLIP and BLIP-2 (multimodal vision-language models), CodeT5 and CodeGen (code-generation models), SFR-Embedding (embedding models), Moirai (time-series foundation model).

Origins

Salesforce was founded in March 1999 by Marc Benioff, Parker Harris, Dave Moellenhoff, and Frank Dominguez to provide cloud-based CRM software. Through the 2000s and 2010s, Salesforce became the dominant global enterprise CRM software provider, with expansion across customer service (Service Cloud), marketing (Marketing Cloud), commerce (Commerce Cloud), analytics (Tableau, acquired 2019), data integration (MuleSoft, acquired 2018), and other enterprise software categories.

Salesforce's investment in AI research began in earnest with the acquisition of MetaMind in April 2016 for approximately $32.8 million. MetaMind was an AI startup co-founded by Richard Socher, the senior NLP researcher and Stanford alumnus who had built one of the earliest deep-learning-based AI products in customer service applications. Socher became Chief Scientist of Salesforce and led the formation of Salesforce Research (renamed Salesforce AI Research in subsequent years), with the principal mandate to apply AI research to Salesforce's product portfolio.

The 2016 to 2020 period under Richard Socher's leadership saw Salesforce AI Research produce research output across natural language processing, computer vision, and other areas. The Einstein AI platform launched in 2016 as the principal commercial-product layer for Salesforce's AI capabilities, integrated across the Salesforce product portfolio. CodeT5 (2021) and CodeGen (2022) established Salesforce's open-source contribution to code-generation research, and the BLIP (2022) and BLIP-2 (2023) multimodal models contributed to the broader vision-language research ecosystem.

In 2020, Richard Socher departed Salesforce to found You.com, the AI-powered search engine. Silvio Savarese, who had joined Salesforce in 2018 from Stanford as Senior Vice President and Chief Scientist of Salesforce AI Research, assumed senior leadership of the research organization. Under Savarese's leadership, Salesforce AI Research has continued research output and has emphasized the integration of AI research into Salesforce's commercial product portfolio.

The 2023 to 2024 period saw Salesforce AI Research expand its open-source releases including XGen (October 2023), SFR-Embedding (July 2024), xLAM (September 2024), Moirai (November 2024), and other variants. The xLAM (Large Action Models) release was particularly significant, with xLAM 8B as the open-weights research variant and a more performant proprietary model powering Salesforce's Agentforce agentic AI platform.

The 2024 to 2026 period has seen Salesforce shift strategic emphasis from foundation-model research to "system-level AI" and "Enterprise General Intelligence" (EGI). Silvio Savarese has articulated EGI as Salesforce's strategic North Star, with the underlying premise that the meaningful breakthroughs are at the agentic-system level (memory architectures, reasoning engines, API calls, and other infrastructure) rather than at the foundation-model layer. Agentforce launched as the principal commercial agentic-AI platform from Salesforce, with xGen-Sales as the proprietary model and other xLAM-derived capabilities providing the underlying agentic infrastructure.

Mission and strategy

Salesforce AI Research's stated mission is to advance artificial intelligence research and to apply AI research outputs to Salesforce's enterprise software products. The dual research-and-product mission has been remarkably consistent since the MetaMind acquisition, with the research division serving as the principal AI capability source for Salesforce's commercial offerings.

The strategy combines four threads. First, foundation-model research across multiple modalities (text via XGen and xLAM, multimodal via BLIP, code via CodeT5 and CodeGen, embeddings via SFR-Embedding, time-series via Moirai), with selective open-source releases for research-community engagement. Second, the Einstein AI platform as the foundational commercial AI layer integrated across Salesforce products. Third, the Agentforce agentic AI platform as the post-2024 commercial focus, with xGen-Sales and other proprietary models providing the underlying agentic capability. Fourth, the EGI (Enterprise General Intelligence) strategic framing, emphasizing system-level AI rather than pure foundation-model competition.

The competitive premise is that enterprise customers benefit from AI capability deeply integrated with their CRM, sales, service, marketing, and commerce data and workflows, and that Salesforce's existing customer relationships and data position the company uniquely to deliver enterprise AI value. The Agentforce launch positions Salesforce as one of the principal enterprise agentic AI platforms, alongside Microsoft Copilot, Google Workspace AI, and other enterprise AI offerings.

Models and products

• xLAM (Large Action Models). xLAM 8B (September 2024) and other variants. Open-source xLAM-1B available for research. More performant proprietary models power Agentforce.
• xGen-Sales. Proprietary sales-task model powering Agentforce. Closed-weights commercial flagship.
• XGen 7B. October 2023 open-weights language model release.
• SFR-Embedding 2B. July 2024 open-weights embedding model.
• CodeT5 and CodeGen. Open-source code-generation model families; foundational contributions to code-generation research.
• BLIP and BLIP-2. Multimodal vision-language models; widely-used in academic and applied multimodal AI research.
• Moirai. Time-series foundation model released November 2024.
• Einstein AI platform. The commercial AI layer integrated across Salesforce's product portfolio.
• Agentforce. The post-2024 agentic AI platform combining xGen-Sales, xLAM-derived capabilities, and other agentic-AI infrastructure.
• Data Cloud. The data foundation underlying Salesforce's AI capabilities.

The principal commercial channels are direct enterprise sales for Salesforce CRM and other products with Einstein and Agentforce included or as upgrades, partner-and-system-integrator distribution through the Salesforce ecosystem, and AppExchange marketplace distribution.

Benchmarks and standing

Salesforce AI Research's open-source research outputs (CodeT5, CodeGen, BLIP, BLIP-2, XGen, xLAM-1B, SFR-Embedding, Moirai) have been consistently characterized in academic-research and industry coverage as competitive with peer open-source releases. The xLAM family in particular has been widely cited as a leading open-source agentic-AI research contribution.

Salesforce's commercial AI standing is anchored by the Salesforce customer base (over 150,000 enterprise customers globally) and the integration of Einstein and Agentforce across the Salesforce product portfolio. The post-2024 EGI strategic framing positions Salesforce as one of the principal enterprise AI platforms globally.

The company's standing in the global AI research community is anchored on the open-source research-output legacy, the Silvio Savarese senior leadership and Stanford research connection, and the integration of research outputs into commercial product. Salesforce AI Research is consistently characterized as one of the principal enterprise AI research divisions globally.

Leadership

As of April 2026, Salesforce AI Research's senior leadership includes:

• Silvio Savarese, Chief Scientist of Salesforce AI Research. Joined Salesforce in 2018 from Stanford. Concurrent appointment as Stanford computer-science professor. Public face for Salesforce AI Research on technical strategy, EGI framing, and external research engagement.
• Marc Benioff, Chair and Chief Executive Officer of Salesforce. Provides senior corporate strategic leadership for the broader Salesforce organization including the AI research division. Public face for Salesforce on agentic AI strategy and the Agentforce launch.
• Senior research-program leadership across the principal Salesforce AI Research program areas. Specific senior leadership for individual research programs has been variously profiled in industry coverage and Salesforce communications.

The research organization includes senior-research talent across the principal program areas, with researchers regularly publishing at the top AI venues (NeurIPS, ICML, ICLR, ACL, EMNLP, CVPR).

Funding and backers

Salesforce AI Research's capital structure is the publicly listed parent Salesforce. Specific R&D investment figures for the research division are not separately disclosed, but Salesforce's annual research-and-development investment exceeds $5 billion as of fiscal year 2025 to 2026.

Salesforce's market capitalization in 2025 to 2026 has been in the hundreds of billions of dollars range on the New York Stock Exchange. The company's commercial revenue scale and cash flow support continued AI research investment alongside product engineering and commercial operations.

Salesforce Ventures, the company's strategic-investor arm, has positions across the broader AI ecosystem including investments in Anthropic, Cohere, Hugging Face (Series D lead investor in 2023), Mistral AI, and other organizations. The strategic-investor portfolio provides additional research-engagement structure beyond the internal research division.

Industry position

Salesforce AI Research occupies a structurally distinctive position in the global enterprise AI landscape. The combination of the foundational MetaMind acquisition lineage, the open-source research-output portfolio (CodeT5, CodeGen, BLIP, BLIP-2, XGen, xLAM, SFR-Embedding, Moirai), the Einstein and Agentforce commercial platform integration, the Salesforce customer base, and the EGI strategic positioning produces a profile that no other enterprise AI organization matches at the same combination of attributes.

Industry coverage frequently characterizes Salesforce as one of the principal enterprise AI platforms globally, with the Agentforce launch positioning Salesforce particularly strongly in the agentic-AI commercial market. The Salesforce Ventures strategic-investor portfolio provides additional research-and-commercial-engagement structure across the broader AI ecosystem.

Strategic risks include intensifying competition from cloud-platform AI services (AWS, Azure, Google Cloud), from frontier-AI labs expanding enterprise capabilities, and from peer enterprise software companies (Microsoft Copilot, Oracle, SAP) deploying agentic AI. Strategic strengths include the Salesforce customer-base depth and integration moat, the open-source research-output legacy, the Silvio Savarese senior research leadership, and the post-2024 agentic-AI commercial positioning.

Competitive landscape

Salesforce AI Research competes with several enterprise AI organizations:

• Microsoft AI (Copilot). Direct enterprise AI competitor through Microsoft 365 Copilot and other products.
• Databricks (Mosaic AI). Direct enterprise AI competitor through Mosaic AI and the Lakehouse platform.
• Snowflake AI (Cortex). Direct enterprise AI competitor through Cortex and other capabilities.
• Google DeepMind, OpenAI, Anthropic, Cohere. Frontier-AI labs whose enterprise APIs compete with Salesforce's foundation-model capabilities. Salesforce Ventures is also a strategic investor in several of these organizations.
• Oracle, SAP, ServiceNow, Workday. Enterprise software peers deploying AI capabilities across their respective product portfolios.
• NVIDIA Research, Together AI. AI infrastructure providers; less direct overlap given Salesforce's enterprise CRM focus versus the infrastructure-platform focus.

Outlook

Several open questions affect Salesforce AI Research's trajectory in 2026 and 2027:

• The continued expansion of the Agentforce commercial platform and the agentic-AI revenue trajectory.
• The integration of Salesforce AI Research outputs (xLAM, xGen, and successors) into Agentforce and other commercial products.
• The trajectory of the EGI (Enterprise General Intelligence) strategic framing and any specific product or platform launches associated with the framing.
• Continued open-source research releases across the foundation-model, multimodal, code, embeddings, time-series, and agentic-AI program areas.
• The competitive dynamic with Microsoft Copilot, Databricks Mosaic AI, Snowflake Cortex, and other enterprise AI platforms.
• The Salesforce Ventures strategic-investor portfolio evolution.
• Continued senior research-talent recruitment and retention.

Sources

• Salesforce AI Research official site. Research division overview.
• Salesforce: Salesforce Delivers Next-Generation AI Models to Power Agentforce. xGen-Sales and xLAM announcement.
• Salesforce Ben: Salesforce Introduces xGen-Sales and xLAM Models. Model technical analysis.
• Salesforce Ben: EGI Is Our True North Star, What's Coming Next for Salesforce AI?. EGI strategic framing.
• Salesforce: We've Reached Peak LLM, Here's What the Next Phase of AI Looks Like. System-level AI strategic articulation.
• Computer Weekly: Salesforce shifts focus from AI models to agentic AI. Strategic-shift coverage.
• Salesforce on Hugging Face. Open-source model distribution.