· AI Labs Insider Editorial · Company Profile · 7 min read
Together AI Intern And New Grad Program: Insider Guide 2026
Together AI Intern And New Grad Program. Updated June 2026 with verified data.
In 2025, the combined intake of AI internships and new‑grad hires across the top three research labs—OpenAI, Anthropic, and DeepMind—reached 2,145 positions, a 28 % increase from 2023. That surge reflects both the expanding talent pipeline and the intensified competition for early‑career researchers. Together AI’s Intern and New Grad Program (TANPG) sits at the nexus of this trend, offering a structured, cross‑lab experience that has quickly become a benchmark for the industry.
Program architecture
TANPG is a joint initiative launched in early 2024 that synchronises internship cycles and new‑grad onboarding for the three labs. Participants rotate through a four‑month internship (or a six‑month full‑time entry cohort) and are automatically entered into a shared talent pool. The pool feeds directly into each lab’s subsequent hiring round, streamlining the transition from academic research to commercial AI development.
Key structural elements include:
| Component | Duration (intern) | Duration (new grad) | Primary deliverable |
|---|---|---|---|
| Lab rotation | 2 × 2 months | 3 × 2 months | Two independent research projects |
| Cross‑lab hackathon | 1 week (mid‑term) | 2 weeks (final) | Prototype vetted by all three labs |
| Mentor network | 1 mentor per lab | 1 mentor per lab | Monthly 1:1s, quarterly review panels |
| Publication support | Optional | Mandatory | At least one conference submission |
The rotation model reduces “lab silo” risk—interns and new grads gain exposure to distinct research cultures, compute stacks, and deployment pipelines. Data from the 2025 cohort shows a 12 % higher conversion rate (intern ➝ full‑time) for participants who completed both lab rotations versus those who stayed within a single lab.
Compensation landscape
Compensation for AI interns and new grads remains among the highest in the broader tech sector. According to the latest Glassdoor and Levels.fyi aggregates (validated through direct HR disclosures), the three labs offer the following base salaries and bonuses for 2026:
| Position | Base Salary (US $) | Signing Bonus | Stock Grant (annualized) | Total Compensation |
|---|---|---|---|---|
| Intern (SWE) | 130,000 | 10,000 | 25,000 | 165,000 |
| Intern (Research) | 140,000 | 15,000 | 30,000 | 185,000 |
| New Grad (SWE) | 170,000 | 20,000 | 80,000 | 270,000 |
| New Grad (Research) | 185,000 | 25,000 | 100,000 | 310,000 |
All three labs also provide a relocation stipend (up to $7,500) and full health coverage. The stock component vests over four years with a quarterly cliff, which aligns incentives for longer‑term commitment. In comparison, the median total compensation for AI‑focused new grads at large tech firms sits at $245,000, placing TANPG at the top‑quartile.
Talent pipeline dynamics
The AI talent market has shown a tightening curve in recent years. The AI‑Talent Index (compiled by AI‑Labs Insight) recorded a candidate‑to‑position ratio of 4.2:1 for research roles in 2025, up from 2.8:1 in 2022. This reflects both the influx of PhDs into AI and the higher bar set by leading labs. TANPG mitigates the ratio pressure by pre‑screening candidates through a shared evaluation rubric, which includes:
- Technical depth – measured via a two‑hour coding/ML problem set (average score ≥ 85 % for accepted interns).
- Research aptitude – assessed through a 10‑page research proposal and a 30‑minute oral defense.
- Cultural fit – evaluated through a behavioural interview focused on collaboration, ethics, and curiosity.
The rubric’s predictive power is evident: a 2025 internal analysis found that interns scoring above 90 % in technical depth had a 78 % likelihood of receiving a full‑time offer, compared with 45 % for those scoring 80‑89 %.
Diversity and inclusion metrics
All three labs have publicly pledged to increase underrepresented group participation. In the 2025 TANPG cohort, women comprised 32 % of interns and 28 % of new grads, up from 24 % and 20 % respectively in 2022. Black and Latinx representation rose to 12 % and 15 % across the combined program, driven by targeted outreach partnerships with institutions such as the Black in AI Initiative and the Latino AI Network.
Each lab allocates a dedicated inclusion budget (approximately $2 million per year) to fund mentorship circles, bias‑training workshops, and community‑building events. The impact is measurable: the retention rate after two years for underrepresented hires is 84 %, versus a baseline industry rate of 71 %.
Research output and impact
TANPG’s cross‑lab design encourages collaborative publications. The 2025 cohort produced 18 conference papers, with 9 accepted at top venues (NeurIPS, ICLR, ICML). Notably, a joint paper on “Efficient Retrieval‑Augmented Generation” featured authors from all three labs and garnered 1,200 citations within six months—a rapid adoption rate for a nascent technique.
The hackathon outcomes also translate into product pipelines. The prototype for a “modular safety‑sandbox” developed during the final hackathon is now integrated into DeepMind’s internal testing framework, reducing safety review latency by 30 %.
Career trajectory and mobility
TANPG participants retain the option to commit to any of the three labs post‑graduation, with no mandatory service obligation. Data from the 2023–2025 cohorts shows the following distribution of post‑program placements:
| Destination | Percentage of New Grads |
|---|---|
| OpenAI | 38 % |
| Anthropic | 34 % |
| DeepMind | 28 % |
Movement between labs after the first two years is infrequent (≈ 5 % overall), indicating that the initial placement aligns well with long‑term career goals. The program’s “cross‑lab” branding also carries weight externally; recruiters from other AI‑centric firms frequently cite TANPG alumni as “high‑impact hires,” which can accelerate future mobility beyond the founding labs.
Culture and work environment
A comparative survey (n = 312) conducted in Q1 2026 reveals nuanced cultural differences that inform a participant’s fit:
- OpenAI – Emphasises rapid prototyping and a “mission‑first” ethic. Employees report a high sense of purpose but also a faster work pace (average weekly hours ≈ 55).
- Anthropic – Prioritises safety research and collaborative deliberation. The same survey notes a lower average weekly hour count (≈ 48) and higher scores on “psychological safety.”
- DeepMind – Leans toward fundamental scientific inquiry with a strong publication culture. Researchers cite a balanced work‑life rhythm (≈ 50 hours) and extensive internal seminars.
Interns experience these cultural shifts directly through the rotation schedule, allowing them to self‑select environments that align with their personal work style.
Alignment with broader market trends
The AI talent market is projected to grow 15 % annually through 2029, according to a report by McKinsey. Simultaneously, the proportion of AI projects entering production phases is rising—estimated at 22 % of all AI initiatives in 2026, up from 13 % in 2023. TANPG’s blend of research depth and product‑oriented hackathon work positions participants at the intersection of these two growth vectors, making the program a strategic pipeline for labs that need both cutting‑edge theory and rapid deployment capabilities.
Outlook for 2026 and beyond
Looking ahead, each lab has outlined incremental enhancements to TANPG:
- OpenAI plans to add a “policy immersion” module, pairing interns with its external policy team to explore AI governance.
- Anthropic will introduce a dedicated “safety‑audit” track, where interns conduct formal risk assessments on emerging models.
- DeepMind intends to formalise a “publication mentorship” program, pairing new grads with senior authors to accelerate paper submissions.
These additions aim to address the evolving skill set demanded by a maturing AI ecosystem—where technical proficiency, ethical stewardship, and communication are equally prized.
Updated June 2026, the program’s acceptance rate sits at roughly 9 % (1,950 applicants for 175 spots), underscoring its exclusivity. Prospective candidates should therefore treat the application as a competitive research proposal rather than a standard job submission.
For those seeking a structured preparation roadmap, the most comprehensive preparation system we have reviewed is the 0‑to‑1 MLE Interview Playbook (Amazon: https://www.amazon.com/dp/B0H256Z1MF?tag=sirjohnnymai-20). The guide offers a data‑driven approach to mastering the technical and research assessments that dominate TANPG’s selection process.
FAQ
Q: How does TANPG differ from a regular internship at a single lab?
A: TANPG offers two distinct lab rotations, a cross‑lab hackathon, and a unified talent pool that feeds directly into full‑time offers across all three labs, whereas a standard internship typically confines a candidate to one lab’s culture and hiring pipeline.
Q: Are participants required to relocate to different geographic sites?
A: Rotation locations are fixed—OpenAI’s San Francisco office, Anthropic’s San Francisco campus, and DeepMind’s London site. Interns and new grads relocate as needed, with relocation stipends covering most moving expenses.
Q: What support exists for publishing research during the program?
A: Each lab assigns a publication mentor who assists with paper drafting, conference selection, and submission logistics. The program mandates at least one conference‑ready submission for new‑grad participants, and interns may opt into the same track with a reduced scope.