llms.blog
AI

Vals AI Raises $40M Series A at $400M Valuation Led by a16z to Build Real-World AI Benchmarks

San Francisco evaluation startup Vals AI announced a $40 million Series A funding round at a $400 million post-money valuation, led by Andreessen Horowitz. The round included participation from existing seed backers 8VC, Pear VC, and Bloomberg Beta, alongside new institutional investors HRT Ventures and Next Ladder Ventures. The financing brings total capital raised by the company to $45 million, following a $5 million seed round. Founded by Stanford computer science graduates Rayan Krishnan a

axc
axc
2 min
Vals AI Raises $40M Series A at $400M Valuation Led by a16z to Build Real-World AI Benchmarks

San Francisco evaluation startup Vals AI announced a $40 million Series A funding round at a $400 million post-money valuation, led by Andreessen Horowitz.

The round included participation from existing seed backers 8VC, Pear VC, and Bloomberg Beta, alongside new institutional investors HRT Ventures and Next Ladder Ventures. The financing brings total capital raised by the company to $45 million, following a $5 million seed round.

Founded by Stanford computer science graduates Rayan Krishnan and Langston Nashold, Vals AI develops automated benchmarking infrastructure designed to grade foundation models and autonomous AI agents on complex, domain-specific enterprise tasks rather than static academic multiple-choice examinations.

Addressing Benchmark Saturation and Data Contamination

The rapid escalation of frontier model training has created a measurement bottleneck across the AI industry. Standard academic benchmarks, such as MMLU or GSM8K, frequently suffer from test set contamination, benchmark saturation, or narrow question formats that fail to reflect production failure modes.

Vals AI addresses this by pairing domain specialists in law, finance, healthcare, and software engineering with automated scoring engines. To prevent models from overfitting or memorizing test sets, the company maintains private evaluation suites that are periodically retired when top-tier models achieve saturation. In May, for example, the firm retired its CorpFin corporate finance benchmark in favor of a dynamic Excel-modeling evaluation suite once frontier systems consistently maxed out the initial test criteria.

Vals AI Benchmarking Architecture and Custom Repository Evaluations

Commercial Expansion, Tool Releases, and Safety Indices

Alongside the Series A financing, Vals AI launched three core products to expand its testing ecosystem:

  1. Vals Smith: A generally available developer tool that automatically constructs custom coding benchmarks from any GitHub repository, allowing engineering teams to evaluate model accuracy against their proprietary codebases and dependencies.
  2. Frontier Risk Benchmarks: A security and safety suite that includes the RSI Index developed in collaboration with CoreWeave, an academic-partnered reverse-engineering cyber evaluation harness, and testing protocols for mental health applications.
  3. Vals Index 2.0: A revamped benchmarking portal and macroeconomic index tracking model capability across enterprise sectors.

According to the company, evaluation results from Vals AI have been incorporated into commercial model cards published by OpenAI, Anthropic, Google, Meta, and xAI. Enterprise customers utilize the platform to select foundation models for production workloads, while policymakers in the U.S. Department of Commerce and Congress have referenced the firm's findings for AI risk assessments.

Vals AI reported an eightfold increase in revenue relative to all of 2025, alongside a doubled customer base and a tripling of its engineering headcount over the past six months. The technical team includes former engineers and researchers from Palantir, Microsoft, Nvidia, Meta, and Hudson River Trading.

Sources

AILLMBenchmarksEvaluationsFunding
ShareX (Twitter)LinkedIn

Written by

axc

More to read

  • Weight-Decomposed Low-Rank Adaptation (DoRA): How Decoupling Magnitude and Direction Closes the LoRA Gap

    Weight-Decomposed Low-Rank Adaptation (DoRA): How Decoupling Magnitude and Direction Closes the LoRA Gap Parameter-efficient fine-tuning (PEFT) has become the standard operational paradigm for adapting large language models to domain-specific downstream tasks. Among existing PEFT methodologies, Low-Rank Adaptation (LoRA) remains the default implementation across industry and academia due to its minimal parameter footprint and zero inference overhead. However, empirical studies consistently reve

    1 min

  • Dynamic KV Cache Eviction in Production: Architecture, Sparsity Policies, and Serving Trade-Offs

    In long-context large language model serving, the key-value (KV) cache is the primary hardware bottleneck limiting concurrency and throughput. While model weights remain static during inference, KV cache memory scales linearly with sequence length, batch size, and layer count. For modern 70B parameter models utilizing Grouped-Query Attention (GQA), serving a 128,000-token context across a modest batch size of 4 requires over 80 GB of VRAM solely for KV states in 16-bit precision, exceeding the m

    1 min

  • Attention with Linear Biases (ALiBi): How Static Positional Slopes Enable Zero-Shot Context Extrapolation

    Large language models process sequences by transforming discrete tokens into continuous vector representations. Standard dot-product self-attention is permutation-invariant: without explicit positional information, the attention operation treats a sequence as an unordered bag of tokens. Early transformer architectures addressed this limitation using Absolute Positional Embeddings (APE), either through fixed sinusoidal functions or learned lookup tables added directly to token embeddings. While

    1 min