SandboxAQ Large Quantitative Models

Learning Objectives

• Understand what Large Quantitative Models (LQMs) are and how they differ from large language models
• Identify the scientific workflows where LQM-via-Claude access changes the procurement and infrastructure picture
• Evaluate the integration against alternative routes to AI-assisted drug discovery and materials science

What Are SandboxAQ Large Quantitative Models?

Large Quantitative Models are simulation-grade numerical models for quantum chemistry, molecular dynamics, and microkinetics — purpose-built physics and chemistry simulators rather than language models. Where an LLM predicts the next token of text, an LQM solves the equations that govern how atoms and molecules actually behave. SandboxAQ — a quantitative-AI company spun out of Alphabet in 2022 and led by CEO Jack Hidary — develops the LQM stack and shipped a flagship integration that exposes its LQMs directly inside Claude in May 2026.

The integration lets computational and research scientists at pharmaceutical and materials companies drive simulation-grade physics models in natural language, without standing up their own high-performance-computing infrastructure. SandboxAQ general manager of AI simulation Nadia Harhen described the launch as "the first time we have a frontier quantitative model on a frontier large language model that someone can access in natural language." Anthropic has not yet detailed the underlying integration mechanism — whether the LQMs are exposed through the Model Context Protocol, a managed API, or an embedded extension.

Pricing Tiers

SandboxAQ has not disclosed standalone LQM pricing for the Claude-integrated path; access is gated through an existing Claude subscription. Enterprise customers should expect bespoke arrangements that cover both the Claude seat and the underlying LQM compute. Direct SandboxAQ commercial channels remain available for customers who need LQM access outside the Claude integration.

Core Capabilities

Quantum Chemistry

LQMs solve the electronic-structure problem — calculating how electrons distribute across a molecule — at scales relevant for drug design and materials screening. Computational chemists can request a binding-energy calculation, a transition-state search, or a reaction-pathway exploration through Claude rather than scripting against a quantum-chemistry package directly.

Molecular Dynamics

The dynamics models simulate how molecules move and interact over time — critical for protein-folding analysis, ligand binding, and materials behavior under stress. The Claude interface accepts a system description in natural language and returns the trajectory or aggregate observables, abstracting away the force-field selection and time-step tuning that traditionally required a specialist.

Microkinetics

Microkinetic models describe how chemical reactions unfold at the molecular level — essential for catalyst design, battery chemistry, and industrial chemical engineering. The integration brings reaction-network analysis into the same conversational interface used for everything else, reducing the path from question to numerical answer.

Natural-Language Driving

The combined effect of LQM access through Claude is that the scientist drives chemistry-grade compute the same way they drive a literature search — by asking. The LQMs handle the numerical work; Claude handles the interpretation, error analysis, and follow-up.

Strengths

• Simulation-grade accuracy: LQMs run deterministic physics, not statistical token prediction — appropriate for regulated drug-discovery and materials work where hallucinations are unacceptable
• No HPC procurement overhead: Access through an existing Claude seat rather than provisioning quantum-chemistry compute infrastructure
• Conversational driving: Natural-language interface for tasks that historically required specialist scripting and force-field expertise
• Unified workflow: Literature search, hypothesis generation, simulation execution, and result interpretation in the same Claude conversation

Limitations and Considerations

• Integration mechanism not yet published: Anthropic has not detailed whether LQMs are exposed via MCP, a managed API, or a custom extension — which affects extensibility and audit posture
• Early-stage Claude integration: Shipped May 2026; production deployment patterns, error-handling conventions, and rate-limit behavior are still being established
• Not a substitute for specialist judgment: LQMs accelerate the compute step, but interpreting catalysis-design or drug-binding results still requires domain expertise
• Pricing visibility: Standalone LQM-via-Claude pricing has not been disclosed; enterprise customers need direct quotes
• Adjacent SandboxAQ products outside scope: SandboxAQ's AQtive Guard (post-quantum cryptography) and quantitative finance offerings are separate product lines

Best Use Cases

When to choose alternatives:

• Protein-structure prediction → AlphaFold 3 / Boltz-2 (problem-specific deep-learning models)
• Free-energy perturbation calculations → Schrödinger Maestro, OpenMM (specialist scientific software with tighter community)
• Phenotypic-discovery imaging at scale → Recursion Pharmaceuticals platform
• Already standardized on a quantum-chemistry vendor → continue with the existing toolchain unless conversational driving is a primary requirement

Getting Started

1. Confirm you have access to Claude Pro, Team, or Enterprise — the LQM integration runs inside Claude
2. Contact SandboxAQ directly for enterprise-scale LQM access details and any tier-specific compute limits
3. Start with a small reference problem — for example a binding-energy or reaction-pathway calculation you already have a baseline number for, to calibrate output quality
4. Build the LQM step into a Claude conversation alongside the literature-search and hypothesis-generation steps you already run there
5. Track which LQM calls survive specialist review — the same accuracy-versus-cost trade-offs that apply to any computational chemistry result still apply here

Strategic Context

The SandboxAQ + Claude integration sits at the intersection of two trends that pharmaceutical and materials companies are watching closely: AI-assisted drug discovery (where AlphaFold's success has set expectations for what AI plus simulation can do) and conversational-AI procurement (where Claude and similar interfaces are becoming the universal scientist workbench). By selling LQM access against existing Claude seats, SandboxAQ converts a specialist-tool procurement decision into a marginal seat-level decision — a meaningful adoption-friction reduction inside large research organizations.

The integration also signals a broader pattern: frontier LLMs becoming distribution channels for specialist scientific compute. The same model — Claude as the workbench, third-party numerical engines as the muscle — is straightforwardly portable to fluid dynamics, structural engineering, electromagnetics, and other quantitative-science domains. SandboxAQ's launch is one of the first published examples.

Key Takeaways

• Large Quantitative Models are simulation-grade physics and chemistry models, not language models — they solve numerical equations rather than predict tokens
• The May 2026 Claude integration lets scientists drive quantum chemistry, molecular dynamics, and microkinetics through natural language, removing HPC procurement from the critical path for pharmaceutical and materials work
• SandboxAQ was spun out of Alphabet in 2022, is headquartered in Palo Alto, and is led by CEO Jack Hidary
• Best fit for drug discovery, catalyst design, battery materials, and any chemistry-grade workflow where the bottleneck has been infrastructure setup rather than scientific judgment
• The integration mechanism with Claude has not yet been fully detailed; early-stage production deployment patterns are still being established