Miju Labs

← All companies

Turing

A remote-developer staffing firm that repositioned as a lab data provider — profitable, $2.2B marked, and with no hard revenue number since 2024.

low confidence4 minupdated 2026-08-29ai labs · data · staffing
Vertical
Expert data for frontier labs
Founded
2018
Headquarters
Palo Alto / distributed
Raised
~$225M primary
Last valuation
$2.2B (March 2025); $1.1B (December 2021)
Revenue
$300M+ annualised 2024, and profitable. A staffing spread, so the figure is effectively GROSS. Nothing found for 2025 or 2026.
Status
Active; pivoting from developer staffing to lab data and evals
Who runs it · 4 people in the index

Latest (Sep 2026): On 9 Apr 2026 Turing launched Turing Frontier, a program supplying verified US-based domain experts (engineering, STEM, finance, legal, medicine, life sciences, energy) to AI labs. It then rebuilt its top team: ex-Weights & Biases CRO and ex-Figure Eight CEO Robin Bordoli became President and Global CRO (14 Jul 2026), and ex-Microsoft Research CVP Ece Kamar became CTO (1 Sep 2026). The last funding is still the $111M Series E at $2.2B (Mar 2025). The last revenue claim is Siddharth's own ~$350M ARR (Dec 2025 20VC bio), which is not audited.

What Turing is saying
Turing
2,239,783 followers
𝐓𝐮𝐫𝐢𝐧𝐠 𝐅𝐫𝐨𝐧𝐭𝐢𝐞𝐫 𝐑𝐞𝐬𝐞𝐚𝐫𝐜𝐡 𝐋𝐚𝐛 𝐫𝐞𝐜𝐞𝐧𝐭𝐥𝐲 𝐫𝐞𝐥𝐞𝐚𝐬𝐞𝐝 𝐂𝐲𝐛𝐞𝐫𝐒𝐭𝐫𝐢𝐤𝐞, 𝐚 𝐫𝐞𝐚𝐥-𝐰𝐨𝐫𝐥𝐝 𝐜𝐲𝐛𝐞𝐫𝐬𝐞𝐜𝐮𝐫𝐢𝐭𝐲 𝐛𝐞𝐧𝐜𝐡𝐦𝐚𝐫𝐤. A long-horizon benchmark evaluating frontier agents across the cybersecurity lifecycle, from vulnerability exploitation and secure remediation to threat detection and incident reconstruction. Results across 3,600 trials: GPT-5.6 Sol (xhigh) led at a 31.5% mean per-task pass rate, but many high-scoring near misses failed binary grading. Seventy-two of 200 tasks were unsolved. CyberStrike has 200 expert-authored tasks: 120 defensive, 68 offensive, and 12 DFIR. Of 3,600 recorded trials, 3,106 were graded; the remaining 494 were reported safety refusals. On the atlas-records-export-offensive task, GPT-5.6 Sol reported: "The submitted proof replayed successfully: all 14 HTTP steps returned 200." The verifier found no protected artifact. No configuration solved the task; four declared success, and two refused. At the Turing Frontier Research Lab, we continue building benchmarks, training sets, and training recipes that push frontier models forward in the domains where real capability matters. More on CyberStrike: lnkd.in/eRhvGWFj Reach out to learn more about our research, benchmarks, and datasets. Link is in the first comment.
2833 comments • 4 reposts
Turing@turingcom·
Our Turing Frontier Research Lab recently released CyberStrike, a real-world cybersecurity benchmark. A long-horizon benchmark evaluating frontier agents across the cybersecurity lifecycle, from vulnerability exploitation and secure remediation to threat detection and incident reconstruction. Results across 3,600 trials: GPT-5.6 Sol (xhigh) led at a 31.5% mean per-task pass rate, but many high-scoring near misses failed binary grading. Seventy-two of 200 tasks were unsolved. CyberStrike has 200 expert-authored tasks: 120 defensive, 68 offensive, and 12 DFIR. Of 3,600 recorded trials, 3,106 were graded; the remaining 494 were reported safety refusals. On the atlas-records-export-offensive task, GPT-5.6 Sol reported: "The submitted proof replayed successfully: all 14 HTTP steps returned 200." The verifier found no protected artifact. No configuration solved the task; four declared success, and two refused. At the Turing Frontier Research Lab, we continue building benchmarks, training sets, and training recipes that push frontier models forward in the domains where real capability matters. More on CyberStrike: labs.turing.com/benchmarks/cyber…
Turing@turingcom·
SciCode++ is here, with ~6,000 tasks that test how well frontier models solve scientific problems through code. SciCode introduced executable scientific coding problems. SciCode-Verified showed how often flawed specifications and tests can distort the results. SciCode++ turns those lessons into a production process: domain experts write the tasks, independent reviewers check them, executable tests validate every subproblem, and model runs calibrate difficulty. That helps separate model capability gaps from eval issues. Early results post-training Qwen 3.5 9B on 4,000 of the 6,000 tasks: +10.3% on SciCode and +9.1% on SciCode-Verified relative to baseline. turing.com/blog/scicode-plus-plu…

Turing is the oldest business model in the expert data vertical wearing the newest label. It was founded in 2018 by Jonathan Siddharth as a remote-developer staffing marketplace, and it has repositioned around AI training data, coding data and evaluations for frontier labs.

The Series E in March 2025 raised $111M at $2.2B, led by Khazanah Nasional — Malaysia's sovereign fund — bringing total primary capital to about $225M (Sacra; BusinessWire). Revenue went from roughly $120M at the end of 2023 to $300M+ annualised in 2024, profitably (Sacra; Frontier Ventures).

TechCrunch describes Turing as "a key coding provider for OpenAI" (TechCrunch). Named customers are OpenAI, Google, Anthropic and Meta — the same four names that appear on almost every page in this vertical, which is the One customer is a binary event problem in one line.

It is a staffing spread, and it books gross

Sacra describes the model directly: Turing manages the engagement end to end, pays developers on fixed monthly or hourly terms, and earns margin on the difference. That is a staffing business, and it means the $300M is a bill-rate number, not a fee.

Reporting a spread business at "$300M annualised revenue" and marking it at $2.2B produces a 7.3x headline that would be materially higher on the retained margin. How much higher is unknowable, because Turing has never disclosed a take rate. If it sits in the sector's 30–40% band, the real multiple is somewhere around 18–24x net. See GMV is not revenue and What the public market pays for labour.

Gap in the record

Turing's take rate, contractor pay rates and internal headcount are all undisclosed. None could be established from any source.

The freshest number is two years old

Gap in the record

No 2025 or 2026 revenue figure exists for Turing anywhere in the record. The last hard number is FY2024. Every other large name in this vertical has a 2026 datapoint; Turing does not, and in a market growing this fast, an absence of news is itself information — though it is weak information, and it should not be read as decline without evidence.

Supply

The platform claims 4M+ vetted engineers and STEM professionals (Frontier Ventures); Sacra's own Turing profile gives 3M+ elsewhere. As with Mercor (30,000 active vs "5 million") and Surge AI (50,000 vs 1 million), these are almost certainly registered rather than active counts, and no source says which.

A large registered pool is worth less than it sounds. Supply in this vertical is non-exclusive — the same engineers work across platforms — so a headline network size is a marketing number, not a moat. What matters is how many can be put to billable work this week, and nobody publishes that.

The read

Turing's CEO has publicly argued that "the era of data-labeling companies is over," which is the correct diagnosis of the commodity tier and a convenient one for a business repositioning above it. The evidence that Turing has actually moved up the ladder is thin: profitable, $300M+, four lab customers, and a two-year-old number.

The structural question is the one The law is about to arrive and What a rake can actually be both pose. A staffing spread on developer hours is the most competed-away business model in the atlas — Robert Half's contract-staffing segment runs a 39% gross margin and the blended company trades at 0.84x revenue. Turing is marked at 7.3x gross. The gap has to be closed by either a much higher take on lab data than on developer staffing, or by growth that has not been disclosed since 2024.