Turing
Turing supplies frontier AI labs with off-the-shelf datasets, reinforcement learning environments and PhD-level experts, and equips large enterprises with embedded engineers and an agent control plane. Access runs through sales contact only, never self-service.
What is Turing?
Turing presents itself as the research accelerator for AI labs and the proprietary intelligence partner for enterprises, under the banner of accelerating superintelligence to drive real economic progress. Run by Turing Enterprises, Inc. from San Francisco, it is not a model and not an assistant: it sells the data, the environments, the people and the engineering that other organisations use to build and evaluate AI. Three domains are targeted: software engineering, enterprise knowledge work and STEM.
The offer splits in two. Frontier AI serves the labs through four capabilities. Off-the-shelf datasets ship with RL environments, tasks, verifiers and traces rather than raw data alone, and cover audio, images, video, molecular structures and Lidar or 3D point clouds. Reinforcement learning environments clone real interfaces such as Jira, Salesforce and Zendesk, add MCP backend environments with APIs, policies, schemas and realistic test data, and capture trajectory traces automatically; they run in three modes, browser, full desktop, or tools and APIs with no interface at all. An expert network of more than five million people across 250+ domains spans finance, software, STEM, law and public policy, medicine, arts and media, vetted in three layers: profile screening, automated testing over 18+ languages with RLHF and SFT tasks and an adaptive AI interview, then in-depth human review. Original research completes the set.
Enterprise AI serves large companies. Forward-deployed engineers recruited from AI labs embed in the client's organisation, code and sprints, alongside the Turing Intelligence Platform, a model-agnostic and cloud-agnostic control plane whose agents run on the client's own infrastructure. Published use cases cover asset management and private equity, banking voice agents and insurance claims.
The Turing Frontier Research Lab publishes original benchmarks, including Turing Terminal-Bench Data (300 tasks, July 2026) and Frontier Agentic STEM (201 tasks, July 2026), and a named catalogue with volumes: SWEBench Public 11,043, HLE++ 10,000, computer-use 3,117, MLE-Bench++ 2,305, SciCode 1,008, CompanyBench 64. Published datasets are distributed on the Turing Data Hub and on Hugging Face. The site claims almost all major frontier AI labs as clients, including OpenAI, Anthropic, Google DeepMind, Meta and NVIDIA.
What it does
- Request dataset samples, generally delivered by email within 48 hours
- Order custom datasets scoped jointly with Turing research scientists
- Access reinforcement learning environments in browser, desktop or MCP tool mode
- Recruit vetted domain experts against a precise specification: domain, degree, seniority, geography, volume, deadline
- Have AI agents built and operated inside your company by forward-deployed engineers
- Deploy, govern and scale those agents through the Turing Intelligence Platform
- Consult the benchmarks and model leaderboards published by the Turing Frontier Research Lab
When to use Turing / When not to
A quick filter to help you decide if Turing is the right fit.
When to use Turing
- Frontier AI labs that train or evaluate models and need expert-grade data at scale across software engineering, enterprise knowledge work and STEM
- Machine learning teams hill-climbing on specific benchmarks such as SWE-bench, LiveCodeBench, MLE-bench, SciCode or HLE, and teams that need reproducible RL environments to train long-horizon agents
- Large enterprises, including the Fortune 500 accounts Turing claims, that want to move an AI proof of concept into agents actually running in production
- Organisations that require agents to run on their own infrastructure without model or cloud vendor lock-in, and research teams that need vetted domain experts fast (one case study cites 1,700+ experts across 30+ domains in under 48 hours)
- Credentialed professionals on the supply side - PhDs, physicians, lawyers, engineers - looking for paid remote work on AI model training
When not to use Turing
- Individuals, freelancers and budget-constrained SMEs: no price is published anywhere and every path leads to a sales form
- Anyone expecting self-service: there is no online trial, no immediate paid sign-up and no product to open and start using
- Developers looking for an API or a developer portal: no API documentation is published, and there is no iOS or Android application
- Teams that need data for immediate commercial use: sample datasets ship under a research-only licence, with full access and commercial terms left to negotiation
- Under-18s, who cannot open an account under the terms of service, and anyone looking for a model or an assistant: Turing sells data, environments, experts and engineering, not something to prompt
How to use Turing
A typical end-to-end flow, from setup to results.
- Start from the contact form at go.turing.com/contact-us or from a 'Request dataset samples' button: there is no self-service entry point
- Browse the dataset catalogue, where each entry shows environment and task type, volume and frontier model scores; published datasets are also available on the Turing Data Hub and on Hugging Face
- Group every dataset you want, off-the-shelf or custom, into a single request - they are delivered in one send
- Receive the samples by email, usually within 48 hours, with metadata; formats include image folders, CSV and JSON for tabular and text data, and WAV for audio
- Talk to the Turing research scientist who follows up, and negotiate full access or commercial terms beyond the research-only sample licence
- For custom datasets, follow the five steps: joint scoping, system and infrastructure build, expert recruitment, operational execution, then confidentiality and security guarantees
- For RL environments, pick an execution mode: browser, full desktop, or MCP tools and APIs without an interface
- To source experts, state the requirement - domain, degree, seniority, geography, volume, deadline - then either pick the candidates yourself or let Turing choose
- For Enterprise AI, follow four steps: bring the use case, scope and plan it, embed the forward-deployed engineers in your organisation, then implement and scale while Turing keeps operating the agents
- On the expert side, create an account on work.turing.com, pass the adaptive AI video interview and the KYC, identity and AML checks, then take assignments with real-time performance feedback
Pros & Cons
Pros
- Quantified, documented dataset catalogue: every entry publishes its volume and frontier model scores - SWEBench Public 11,043, HLE++ 10,000, computer-use 3,117, MLE-Bench++ 2,305, SciCode 1,008 and more
- Original benchmarks published by the Turing Frontier Research Lab, with downloadable fact sheets
- Named reference customers: the site claims almost all major frontier labs, including OpenAI, Anthropic, Google DeepMind, Meta and NVIDIA
- Depth of the expert network: 5M+ people, 250+ domains, 140+ countries, with a case study delivering 1,700+ vetted experts across 30+ domains in under 48 hours
- Vetting that is documented and configurable per project, including KYC and AML checks
- On Enterprise AI, agents run on the client's own infrastructure, data does not leave the client environment, there is no model or cloud lock-in, and the client keeps the code, the IP and the direction
- Established company: $300M+ raised, a 4.3 Trustpilot rating according to the home page, Forbes, The Information and Fast Company recognition, and a detailed privacy policy with appointed EU and UK GDPR representatives
Cons
- No public pricing whatsoever - no grid, no range, no entry point; everything goes through a contact form
- No self-service, no trial and no immediate sign-up on the customer side
- No API documentation and no developer portal
- No named security certification: neither SOC 2, nor ISO 27001, nor HIPAA appears anywhere on the site, which is notable for a B2B offer aimed at large accounts
- No published subprocessor list, and no hosting country or region stated: data may be stored anywhere in the world
- No documented way to opt out of model training, while the privacy policy declares AI training as a purpose and allows selling personal information to customers for consideration
- A very broad perpetual and irrevocable licence over User Content, terms of service dated 14/03/2024 that predate the Frontier AI / Enterprise AI overhaul, and sample datasets limited to a research-only licence
Pricing & Plans
No price is published on any page of the site and there is no pricing page: every call to action leads to a contact form. Turing operates on a contact-sales basis, with quotations drawn up after a scoping discussion, custom dataset engagements being structured around the client's timing, scope and cost requirements. Sample datasets are provided free of charge but under a research-only licence, full access and commercial use being subject to negotiation. No permanent free plan and no free trial are advertised on the customer side. On the expert side the flow is reversed: Turing pays contributors, with rates displayed on the home page ranging from 150 to 1,000 USD per role and payments made twice a month.
- every engagement is quoted after scoping
- ordered from the published catalogue
- with free samples under a research-only licence
- scoped with Turing research scientists through a five-step engagement
- vetted domain experts recruited against a written specification
- forward-deployed engineers plus the Turing Intelligence Platform
Data, GDPR & hosting
A consolidated view of how Turing handles your data.
GDPR overview
GDPR is addressed concretely, though compliance is never claimed in so many words. The privacy policy, updated 27/06/2025, carries a dedicated Supplemental notice for EU/UK GDPR. Turing Enterprises, Inc. and its affiliates are named as controller. Legal bases are mapped section by section - contract performance, legitimate interest, consent, legal obligation - with specific conditions for special categories of data. Rights are listed: confirmation of processing, identification of recipients, access and portability, rectification, erasure, restriction and objection, withdrawal of consent, and complaint to a supervisory authority in the EEA, the UK or Brazil. Identity verification is required. International transfers are acknowledged, with EU Standard Contractual Clauses cited as a possible safeguard. Article 27 representatives are appointed: Osano in Dublin for the EU and Osano UK in Belfast. GDPR contact: privacy@turing.com. No SOC 2, ISO 27001 or HIPAA certification is named.
Who owns the data?
The terms of service state that Turing claims no ownership over User Content, which remains yours. In exchange you grant Turing a royalty-free, sublicensable, transferable, perpetual, irrevocable, non-exclusive worldwide licence to use, reproduce, modify, publish, translate, distribute and make derivative works of that content and of your name, voice and likeness, with moral rights waived and no compensation. User Content may be treated as non-confidential, may be visible to other users and third-party sites, and may be deleted permanently if your account is terminated. Intellectual property in the Services themselves stays with Turing Enterprises, Inc. and its licensors. Enterprise AI reverses this for deliverables: you own the code, the IP, and the direction. Customer Data processed under a signed DPA falls outside the privacy policy.
Reuse rights
Your own User Content stays yours and you may reuse it freely, without asking Turing. What the documents do not give you is any right over Turing's own material: the Services and their intellectual property remain Turing's, and sample datasets are supplied under a research-only licence - full access or commercial use has to be negotiated. In the other direction, the privacy policy sets out five families of purposes for the data Turing collects, and section 3B explicitly names developing, training, fine tuning, and prompting artificial intelligence models. Collection is broad: IP address, cookie identifiers, browser, approximate and precise geolocation, pages visited, frequency and duration, plus audio and video recordings of training sessions, meetings and live coding interviews (name, image, voice, screen sharing). Data may be disclosed to service providers, to Turing customers, to business partners, affiliates and advertising partners, and Turing states it may sell or otherwise make personal information available to its customers in exchange for monetary or other valuable consideration. Do Not Track signals are not honoured, and no opt-out from model training is documented. Customer Data processed as a processor is carved out and governed by the DPA instead.
Data retention & training
Hosting summary
No hosting country, region or provider is named anywhere on the site. Section 6 of the privacy policy states that all personal information processed by us may be transferred, processed, and stored anywhere in the world, including, but not limited to, the United States or other countries, and warns explicitly that those countries may have data protection laws different from the ones where you live. For transfers from the EEA, Switzerland and the United Kingdom, the EU Standard Contractual Clauses are cited as a safeguard Turing may use. Hosting providers appear among the categories of service providers receiving data, but none is named and no subprocessor list is published. The company and its published address are in the United States, in San Francisco. The resolved IP address geolocates to the United States, but it belongs to an Imperva/Incapsula anycast CDN node and therefore says nothing about where data actually sits. Enterprise AI inverts the logic: agents run on the client's own infrastructure, the data does not leave the client environment, and the choice of cloud belongs to the client.
Things to keep in mind
Risks and trade-offs to weigh before adopting Turing.
- The privacy policy names developing, training, fine tuning, and prompting artificial intelligence models among the purposes for the personal information collected, and no opt-out from that training is documented anywhere
- The same policy states that Turing may sell or otherwise make personal information available to its customers in exchange for monetary or other valuable consideration
- The User Content licence is exceptionally broad: worldwide, perpetual, irrevocable, sublicensable and transferable, with moral rights waived, no compensation, and coverage extending to your name, voice and likeness
- Recordings of technical interviews may be shared with customers for job matching; deletion is possible on request to support@turing.com, at the cost of fewer matching opportunities
- Do Not Track signals are explicitly not honoured, and automatic collection extends to precise geolocation as well as the frequency and duration of visits
- Affiliated Persons - employees, contractors and freelancers - are subject to broad monitoring of devices, applications, accounts and messaging, with no expectation of privacy declared
- Governance gaps to weigh before committing: no named security certification, no subprocessor list, no stated hosting country, terms of service dated 14/03/2024 that predate the current product line, a PMB mailbox as the published postal address, and no public price to benchmark against
Setup & Integrations
Technical difficulty
Low technical difficulty, high commercial friction. Nothing is installed and there is no self-service onboarding: the starting point is a scoping conversation. Off-the-shelf datasets arrive by email, usually within 48 hours, in AI-ready formats that can be ingested directly. Custom datasets follow a five-step process led by Turing research scientists and project managers. RL environments require choosing an execution mode: browser, desktop or MCP tools. For Enterprise AI, Turing engineers embed in the client's team, code and sprints, then operate the agents. The client-side technical load is light; the commitment is a project, not a subscription.
Deployment
Integrations
Behind Turing
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Who is Turing for?
Which AI labs does Turing work with?
How quickly do dataset samples arrive, and in what formats?
What licence covers the sample datasets?
Is there public pricing?
Who makes up the expert network, and how are experts vetted?
Where do Enterprise AI agents run, and who owns the code?
Does Turing train AI models on the data it collects, and can I opt out?
Is there an API or a mobile app?
Where can I find Turing's published datasets and benchmarks?
Should you pick Turing?
Turing is an established B2B supplier in the market for data, environments and evaluation aimed at frontier AI. Its credibility rests on things that can be checked: a dataset catalogue with published volumes and frontier model scores, original benchmarks released by its research lab, an expert network of five million people across 250+ domains and 140+ countries, and named reference customers. On the enterprise side, running agents on the client's own infrastructure, with no model or cloud lock-in and the client keeping the code and the IP, answers the objection most often raised against consultancy-led AI projects.
The friction is commercial rather than technical. There is no public price at all: no grid, no range, no entry point, no self-service. Everything starts with a form and a scoping call, which rules out any budget comparison before engaging and rules out the tool entirely for individuals, freelancers and small teams.
Two things deserve a careful read before signing. First, the privacy policy names the development, training and fine tuning of AI models among the purposes for the personal information collected, states that Turing may sell or otherwise make personal information available to its customers for consideration, and documents no way to opt out of training; the terms of service also take a perpetual, irrevocable, sublicensable licence over User Content, extending to name, voice and likeness. Second, the governance disclosure is thin for an offer aimed at large accounts: no named security certification, no subprocessor list, no hosting country, and terms dated March 2024 that predate the current product line.
Turing suits organisations with a budget, a procurement function and legal review capacity. Ask for the DPA, the security documentation and the commercial dataset licence in the very first conversation.
- Choosing a selection results in a full page refresh.
- Opens in a new window.