Releases

Scale and Google Cloud published a joint architecture for deploying Scale’s GenAI Portfolio on Google Cloud with Gemini Enterprise, identity, governance, and cloud infrastructure.

Scale’s board appointed former Google Cloud and Illumina executive Francis deSouza as chief executive officer, effective August 10, 2026.

Scale partnered with Carahsoft to make its GenAI Platform, test-and-evaluation capabilities, and deployment services available through U.S. government procurement schedules.

Scale marked its tenth anniversary, reporting more than 700 customers served, 15 billion human decisions applied to model development, work across 150 languages and locales, and over $1 billion paid to contributors.

Scale completed SWE Atlas by adding Test Writing and Refactoring alongside Codebase QnA, producing a 284-task software-engineering evaluation suite.

The U.S. defense department increased the potential ceiling of its Scale production agreement from $100 million to $500 million following rapid adoption across military components.

Scale signed a memorandum of understanding with the U.S. Department of Energy to explore AI-ready scientific data infrastructure, evaluations, and model applications for the Genesis Mission.

Scale acquired ICG Solutions and its LUX real-time streaming-data analytics platform to expand its national-security AI stack; ICG initially remained a wholly owned subsidiary.

Scale Labs updated MultiChallenge with a stronger judge model, refined tasks, and refreshed results to improve agreement with expert ratings.

Scale integrated its Physical AI Data Engine with Universal Robots’ AI Trainer to collect production-grade visual and force-feedback data on industrial robots.

Scale Labs introduced repository-grounded rubrics for evaluating and reranking candidate software patches without executing tests.

Scale Labs introduced VeRO, a reproducible harness using versioned agent snapshots, controlled budgets, structured traces, and reference procedures.

Scale expanded SEAL into Scale Labs, a broader research organization focused on evaluation, agents, multimodal systems, post-training, deployments, and oversight.

Scale launched SWE Atlas, initially releasing Codebase QnA to evaluate how coding agents investigate and reason about real software systems.

Scale launched high-fidelity simulated environments for training and testing agents on tool use, computer use, coding, and professional workflows.

Scale researchers released LHAW, a framework for generating underspecified long-horizon tasks and evaluating whether agents clarify ambiguity and recover performance.

Scale reported that 2025 was its strongest financial year, with more than $1 billion in new business, profitable data operations, and accelerating applications and government work.

Qatar’s Ministry of Communications and Information Technology launched the Arabic-first ALIF AI fluency program with Scale AI for government professionals.

Scale announced expansion across New York, London, Washington, D.C., Doha, and St. Louis to support enterprise, government, and international growth.

Scale released MCP Atlas to evaluate how well AI agents combine tools across real Model Context Protocol servers.

Scale expanded the SEAL Leaderboards into agent evaluation, beginning with SWE-Bench Pro and MCP Atlas.

Scale released a harder, contamination-resistant benchmark using real and commercial repositories to measure software agents on complex multi-file engineering tasks.

Scale updated its mission to developing reliable AI systems for the world’s most important decisions, reflecting its expansion from data infrastructure into applications and evaluation.

The U.S. defense department awarded Scale a five-year production agreement with an initial $100 million ceiling for data operations, computer vision, GenAI, and Donovan capabilities.

Scale reduced approximately 200 employee roles and contractor assignments after determining it had expanded its generative-AI capacity too quickly, while prioritizing enterprise, government, and international growth.

Meta made a $14.3 billion strategic investment for a 49% non-voting minority stake in Scale AI, valuing the independent company at over $29 billion and expanding the commercial partnership.

Founder and CEO Alexandr Wang left day-to-day management to join Meta’s AI efforts while remaining a Scale director; Chief Strategy Officer Jason Droege became interim CEO.

Scale formalized a data-engine offering for robotics and embodied AI, covering multimodal collection, annotation, and real-world model improvement.

The U.S. Defense Innovation Unit selected Scale to lead Project Thunderforge with Anduril and Microsoft, integrating AI agents into operational military planning.

Scale entered a five-year partnership with Qatar’s Ministry of Communications and Information Technology to deploy AI applications, analytics, automation, and workforce programs across government.

Scale published MASK as part of its safety research program for testing whether models knowingly state beliefs inconsistent with their internal representations.

Scale released FORTRESS to evaluate frontier-model safeguards against dual-use national-security and public-safety risks.

Scale and the Center for AI Safety published Humanity’s Last Exam, assembled from nearly 1,000 contributors across more than 500 institutions in 50 countries.

Scale added richer model comparison, failure analysis, cross-sectional reporting, and automated error-discovery capabilities to its frontier evaluation framework.

Scale released Defense Llama, a Llama 3-based model fine-tuned for controlled U.S. national-security use cases inside Scale Donovan.

Scale and the Center for AI Safety launched an international effort to create a broad expert-written benchmark for frontier AI systems.

Scale and research collaborators introduced EnigmaEval, a multimodal reasoning benchmark built from novel puzzle-competition problems.

Scale released MultiChallenge to measure model performance on realistic multi-turn conversation problems involving memory, instruction retention, editing, and self-consistency.

Scale made its model-evaluation platform generally available for model developers, enterprises, and public-sector organizations.

Scale launched expert-evaluated leaderboards for coding, instruction following, mathematics, and multilinguality using private, contamination-resistant prompt sets.

Scale closed a $1 billion primary-and-secondary financing led by Accel at a $13.8 billion valuation.

Scale expanded its generative-AI platform and services into custom agentic applications connected to enterprise data, controls, and workflows.

Scale launched its Safety, Evaluations, and Alignment Lab to build standardized frontier-model benchmarks, evaluation products, and red-team methods.

OpenAI named Scale a preferred partner for enterprise fine-tuning of GPT-3.5, combining Scale’s data engine with OpenAI models.

Scale provided evaluation infrastructure for the White House-supported Generative Red Team Challenge at DEF CON 31, testing frontier models with thousands of participants.

Scale introduced an expert-driven testing and evaluation platform for measuring frontier-model capability, safety, bias, and robustness.

Scale launched EGP as a model-agnostic enterprise platform for deploying secure generative-AI applications with proprietary data, fine-tuning, evaluation, and red teaming.

Scale launched Donovan, a secure decision-support platform using large language models, retrieval, and mission data for government and defense users.

Scale deployed Donovan for the U.S. Army XVIII Airborne Corps, describing it as the first large-language-model platform deployed on a classified network for the Corps.

Scale partnered with Anthropic to help enterprises customize and deploy generative-AI applications powered by Claude.