What Vals.ai does

Vals.ai is a San Francisco-based AI evaluation company founded in 2024 that builds independent benchmarking infrastructure for foundation models and agentic AI systems. The company's core product is a SaaS evaluation platform that runs domain-specific benchmarks across finance, legal, healthcare, software engineering, and academic tasks, using private test sets to prevent dataset leakage and expert-authored rubrics to grade performance. The platform supports evaluation of text, code, structured data, and multimodal inputs, and includes agentic capability testing (tool-calling, multi-turn flows, computer-use). Notable benchmarks include the Vals Index (a composite weighted by sectoral contribution to U.S. GDP), Vals Multimodal Index, Legal Research Bench, Harvey's Legal Agent Benchmark, Code Migration, SkillsBench, SWE-bench Verified, Terminal-Bench, and vertical-specific evaluations such as Finance Agent v2, CorpFin v2, MedCode, and MedScribe.

The business model is enterprise SaaS delivered through a self-serve platform with SDK, CLI, and CI/CD integrations for engineering teams, combined with an enterprise sales overlay that provides quote-based pricing, custom evaluations, and private dataset licensing on a case-by-case basis. The company serves AI labs, enterprise engineering teams, and AI application vendors in legal, financial services, healthcare, insurance, and software engineering verticals, with primary buyer pain points centered on selecting and validating AI models for high-stakes production use cases. Vals.ai has earned coverage from Bloomberg, the New York Times, and TechCrunch, achieved SOC 2 Type 2 certification, and partnered with Harvey, Stanford researchers, BenchFlow AI, and industry domain experts to produce benchmarks.

The company was launched with seed funding from Pear VC and operates with a 1-10 employee team. Revenue, headcount growth, and follow-on funding rounds have not been publicly disclosed. The product launch cadence accelerated in 2026 with multiple new benchmark releases, and the company has begun building an open ecosystem through the Valkyrie public benchmark registry.

Name

Vals.ai

Legal name

Vals AI, Inc.

Website

https://vals.ai

Company type

Private

Founded year

2024

Operating status

Operating

Headcount range

11–50 employees

Short description

Vals.ai builds independent AI benchmarking infrastructure that evaluates foundation models and agents on private, domain-specific tasks in finance, legal, healthcare, software, and academia, serving AI labs and enterprise teams via a SaaS platform with SDK and enterprise sales.

Ownership category

akta.pro rank

Product category

AI Model Evaluation Platform

NAICS

Software Publishers (5132), Software Publishers (513210), Other Computer Related Services (541519)

SIC

Services-Computer Programming Services (7371), Services-Computer Integrated Systems Design (7373)

akta.pro primary industry

AI Observability, Monitoring & Evaluation Platforms (Drift, Quality, Safety) (HDAEANAF)

akta.pro secondary industries

Enterprise AI Governance, Risk & Compliance Platforms (Model Risk, Audit, Policies) (HDAEANAE), Audit, Explainability & Accountability Tooling (traceability, reporting) (HDAAAKAL), Responsible AI, AI Governance & Compliance Services (BPAEAHAJ)

Keywords

• AI model benchmarking
• LLM evaluation platform
• Domain-specific AI testing
• Agent performance evaluation
• AI safety benchmarks

Headquarters

HQ city

San Francisco

HQ country

United States

HQ region

North America

Markets served

GTM type

B2B

Offering type

Software

Cost components

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

Revenue model

1. Enterprise Benchmarking Platform: SaaS platform subscription model where enterprises pay for access to evaluation infrastructure, benchmarking tools, and expert review capabilities for assessing AI models and agents in high-value domains.
2. Private Benchmark Licensing: Companies license private validation datasets for their own internal validation, with statistical proof correlated to Vals' test suite.

Pricing tiers

Go-to-market motion2 records

Distribution channels2 records

Marketing channels5 records

Core offering

Vals.ai operates an independent AI model and agent benchmarking platform that evaluates large language models on private, domain-specific tasks spanning finance, legal, healthcare, software engineering, and academia. The platform collects review criteria from subject-matter experts, runs evaluations at scale via SDK/CLI tooling, and publishes indices such as the Vals Index and Vals Multimodal Index. It targets enterprises and AI labs that need unbiased, leakage-resistant performance signals for model selection, regression testing, and deployment decisions.

Product overview

Vals AI operates as a benchmarking and evaluation platform for AI models and agents. The core offering is the Vals AI Platform, an evaluation infrastructure that allows labs and engineering teams to collect review criteria from subject-matter experts, run evaluations at scale, and drive their review process. Built on this platform are industry-specific benchmarks including the Vals Index (aggregated score across domains) and Vals Multimodal Index (finance, coding, education weighted by economic contribution), along with specialized benchmarks for legal research (Legal Research Bench, Harvey's Legal Agent Benchmark), coding (SWE-bench Verified, Vibe Code Bench, Terminal-Bench, ProgramBench, Code Migration), finance (Finance Agent v2, CorpFin v2), and healthcare (MedCode, MedScribe). The company also offers developer tools including an SDK and CLI for CI/CD integration, and maintains the Valkyrie public benchmark registry for community contributions. All benchmarks are designed to be private and secure to prevent dataset leakage, with public validation sets for transparency.

Differentiator

Problem solved

Functional benefit

Products and services

• Vals AI Platform Evaluation infrastructure platform enabling labs and engineering teams to collect review criteria from subject-matter experts, run evaluations of any LLM model at scale, and drive their review process. Supports benchmarking of foundation models and LLM applications on task-specific data.
• Vals Index Aggregated benchmark score combining multiple domain-specific evaluations weighted by economic sector contribution. Top performing models include Claude Fable 5 (75.14%), Claude Opus 4.8 (70.36%), and GPT 5.5 (67.95%).
• Vals Multimodal Index Benchmark measuring AI model performance across finance, coding, and education sectors, weighted by each sector's contribution to the U.S. economy using Federal Reserve Economic Data and Bureau of Labor Statistics.
• Legal Research Bench Private benchmark testing whether agents can handle realistic US legal research across eight practice areas using case-law search, web search, and document retrieval, graded against rubrics authored and peer-reviewed by practicing lawyers.
• Harvey's Legal Agent Benchmark (HLab) Benchmark testing an agent's ability to complete legal work by answering client inquiries using shell and file-editing tools plus skills for Word, Excel, and PowerPoint.
• Code Migration Proprietary benchmark asking whether models can reimplement working programs in another language. Includes CLI split (30 repositories across Python, Java, Kotlin, Rust, and C++) and COBOL-to-Java split.
• SWE-bench Verified Benchmark evaluating software engineering capabilities of AI models on real-world tasks from GitHub repositories.
• Vibe Code Bench Benchmark testing AI model performance on coding tasks, with DeepSeek V4 ranking #1 among open-weight models.
• Terminal-Bench 2.1 Benchmark evaluating terminal and command-line interface capabilities of AI models.
• ProgramBench Benchmark evaluating whether models can reconstruct command-line programs from an executable binary and behavioral specification.
• SkillsBench Benchmark evaluating whether agents improve at software tasks when given reusable, task-specific knowledge. Average score improved from 35.53% without skills to 52.53% with skills.
• Finance Agent v2 Benchmark evaluating AI agent performance on financial tasks and workflows.
• CorpFin v2 Benchmark evaluating AI model performance on corporate finance tasks.
• MedCode Benchmark evaluating AI model performance on medical coding tasks.
• MedScribe Benchmark evaluating AI model performance on medical transcription and documentation tasks.
• Valkyrie Public Benchmark Registry Public benchmark registry and evaluation framework allowing community contribution of benchmarks compatible with Vals AI's evaluation platform. Integrated with SkillsBench.
• Vals AI SDK Python software development kit enabling developers to integrate Vals AI's evaluation infrastructure into their applications for benchmarking LLM performance.
• Vals AI CLI Tools Command-line interface tools for running evaluations, enabling CI/CD integration and automated testing workflows for LLM applications.

Quantifiable outcome

• Claude Fable 5 leads Vals Index at 75.14%, outperforming other publicly available AI technologies by 5% according to Vals AI benchmark tests
• +3 more outcomes

Segments6 records

Ideal customer profiles5 records

Technology focussed Yes

API detail

Has API

Yes

API docs

API detail

Core technology

AI maturity

App detail

AI capability8 records

Feature5 records

Partnerships

Three partnerships are on record, tiered core and secondary.

• Stanford ResearcherscoreStrategic or Co-development PartnerCollaboration with Stanford university researchers to develop domain-specific AI benchmarks and evaluation methodologies.
• Industry ExpertscoreStrategic or Co-development PartnerPartnership with industry experts to create benchmarks that reflect real-world industry use cases across finance, legal, healthcare, and other domains.
• BenchFlow AIsecondaryStrategic or Co-development PartnerPartnered with BenchFlow AI to integrate SkillsBench into Vals AI's Valkyrie evaluation framework, contributing to the public benchmark registry.

Scale indicators3 records

Recent moves6 records

Expansion highlights6 records

Broad incumbents

• Hugging Face: Hugging Face hosts the Open LLM Leaderboard and other community benchmarks, directly competing with Vals.ai on independent model evaluation. It also offers a much broader open-source model and dataset ecosystem, making it a broad incumbent in the same evaluation-adjacent space.
• Scale AI: Scale AI operates a large AI evaluation and data-labeling platform serving foundation model labs and enterprises. It overlaps with Vals.ai on model evaluation and benchmarking services, but at much larger scale and with a broader data-services portfolio.
• Weights & Biases: Weights & Biases offers MLOps tooling including LLM evaluation, experiment tracking and reporting. It is a broad incumbent overlapping with Vals.ai on enterprise evaluation, reporting and audit workflows.
• MLPerf (MLCommons): MLCommons runs industry-standard ML benchmarking suites (MLPerf). While focused more on training/inference systems than domain tasks, it is comparable to Vals.ai as a standardized, independent benchmark authority.
• LangSmith (LangChain): LangSmith is an LLM application evaluation, debugging and monitoring suite bundled with the LangChain ecosystem. It competes with Vals.ai on enterprise LLM evaluation and CI/CD-integrated testing workflows.

Direct peers

• Patronus AI: Patronus AI builds enterprise-focused LLM evaluation and safety testing products. It overlaps directly with Vals.ai on domain-specific evaluation (finance, legal) and agent/system-level assessment for enterprise buyers.
• Braintrust: Braintrust provides an evaluation and observability platform for LLM applications with scoring, datasets and CI/CD integration. It is comparable to Vals.ai on the developer-tooling and regression-testing layer for AI applications.
• Artificial Analysis: Artificial Analysis provides independent benchmarks and comparisons of AI models across quality, latency and cost dimensions. It is the closest direct peer to Vals.ai, with similar positioning around unbiased, third-party model evaluation.
• LMSYS (Chatbot Arena): LMSYS Chatbot Arena is a widely cited crowd-sourced benchmark for comparing LLMs. It competes with Vals.ai as an independent evaluation channel cited by major publications and used by labs to validate model launches.

Emerging players

• Stanford CRFM (HELM): Stanford CRFM's HELM benchmark is a leading academic, transparent evaluation of foundation models. It is comparable to Vals.ai as a transparent, multidimensional benchmark, though it operates as a research effort rather than a commercial product.

Market position

Strengths4 records

Weaknesses4 records

Competitive moat4 records

Key risks5 records

Key highlights7 records

Customer concentration

• Xx.com/_valsai
• LinkedInlinkedin.com/company/vals-ai

Compliance1 record

Revenue estimate

Valuation estimate

Number of profiles

Funding overview

Funding rounds2 records

Investors6 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 Vals.ai