Meta Superintelligence Labs shipped its first proprietary foundation model, Muse Spark, on April 8, 2026 — roughly nine months after Mark Zuckerberg announced the lab's creation in a memo dated June 30, 2025. Behind that model is a ~50-person research unit called TBD Lab, which Meta assembled through the most concentrated talent-acquisition campaign in US AI history. In H1 2026, TBD Lab's recruiting pipeline — not its published research — is the clearest signal of where Meta's AGI strategy actually sits.
What Happened: The Build
On June 30, 2025, Zuckerberg's internal memo announced the formation of Meta Superintelligence Labs (MSL), a consolidated unit placing all of Meta's AI research under a single structure for the first time. The memo named former Scale AI CEO Alexandr Wang as Chief AI Officer and former GitHub CEO Nat Friedman as co-leader. By August 2025, MSL had been organized into four operating units: TBD Lab, focused on frontier model development; FAIR, the legacy fundamental research group; a products and applied research arm under Friedman; and MSL Infra under VP of Engineering Aparna Ramani.
TBD Lab is the unit that matters for this analysis. At approximately 50 researchers and engineers as of Q2 2026, it is the smallest of the four groups by headcount and the only one that Zuckerberg characterized publicly as operating outside normal Meta management norms. In September 2025, he told the State of AI podcast with Rowan Cheung that TBD Lab runs as "a group science project" with a flat leadership structure and no C-suite deadlines. "Everyone's competitive," Zuckerberg said. "They all want to be at the frontier and doing leading work. Me setting a deadline for them isn't going to help."
The sourcing record for those 50 seats tells a more structured story than the "science project" framing suggests. Meta recruited more than 20 researchers from OpenAI between June and December 2025 [ENTRA estimate based on named individual hires reported by TechCrunch, CNBC, and WSJ through December 2025; Zuckerberg's June 30, 2025 memo named 10 initial hires], including Shengjia Zhao — a co-creator of ChatGPT — who was named Chief Scientist of MSL in July 2025 (per CNBC, July 25, 2025). From Google DeepMind, Meta pulled Jack Rae (pre-training tech lead for Gemini 2.5), Pei Sun (perception models, formerly Waymo/DeepMind), Tong He, and Yuanzhong Xu. From Apple, Ke Yang, who led the machine learning effort behind Apple's ChatGPT-comparable search experience, joined in October 2025. From Anthropic, Joel Pobar, who worked on inference infrastructure, crossed to Meta in the same period.
The highest-profile acquisition came from Thinking Machines Lab, Mira Murati's post-OpenAI startup. Andrew Tulloch, a co-founder, joined MSL in late 2025 with a compensation package reported by the Wall Street Journal and subsequently covered by The Next Web and multiple other outlets at up to $1.5 billion over at least six years, including performance bonuses and stock incentives. A Meta spokesman called that characterization "inaccurate and ridiculous" — without disclosing an alternative figure. In March 2026, Meta completed a second Thinking Machines Lab acquisition: founding engineer Joshua Gross joined MSL as part of a broader talent raid in which five of the startup's founders were absorbed across Meta and competitors.
Additionally, in March 2026 Meta acquired Moltbook, an AI agent-only social network, with founders Matt Schlicht and Ben Parr joining MSL as part of the acqui-hire, effective March 16.
The New Comp Tier: What the Numbers Say
The compensation architecture MSL deployed in H1 2026 created a new senior IC bracket in the US AI labor market that did not exist in 2024. It operates in three layers.
Layer 1: Posted bands (base only). Meta's H-1B visa filings — public record — show AI Research Scientist roles with base salaries between $163,800 and $328,000. A VP of AI position was filed at $650,000 base. These figures exclude equity, which for senior MSL hires represents the majority of realized value.
Layer 2: Standard senior research comp. Levels.fyi data through May 2026 shows Meta Research Scientist total compensation ranging from $305,000 at IC4 to $581,000 at IC6. The IC6 band breaks down to approximately $217,000 base, $313,000 in annual RSU value on a four-year vest, and $50,600 in bonus. Those figures represent a material step up from the 2024 baselines on the same platform.
Layer 3: TBD Lab bespoke packages. Above IC6, and for the lateral hires Meta has targeted from competing frontier labs, the structure moves outside published bands entirely. Packages reported for this cohort range from $100 million signing bonuses (claimed by Sam Altman in public remarks on June 17–18, 2025, per Bloomberg and CNBC, and framed as evidence of market escalation; disputed by Meta CTO Andrew Bosworth and by named researcher Lucas Beyer, who called the figure "fake news" — the structure appears to involve total packages, not pure signing bonuses) to the $200 million total package Ruoming Pang reportedly received in mid-2025 to leave Apple and join MSL (per Bloomberg, July 9, 2025). The upper bound — the Tulloch figure, disputed or not — sits at nine figures over six years.
| Level | Base (approx.) | Annual RSU | Bonus | Total Comp | |-------|---------------|------------|-------|------------| | IC4 Research Scientist | $158K | $120K | $27.5K | $305K | | IC5 Research Scientist | $202K | $128K | $25.6K | $356K | | IC6 Research Scientist | $217K | $313K | $50.6K | $581K | | TBD Lab bespoke (senior lateral) | Undisclosed | Eight- to nine-figure packages | — | $100M–$300M+ reported range |
Sources: Levels.fyi (IC4–IC6, May 2026); Meta H-1B public filings; The Next Web, Fortune, TechCrunch (bespoke packages).
The gap between IC6 and the bespoke tier is not a comp band — it is a separate market. It requires Zuckerberg's personal involvement in recruiting, which multiple reports confirm he has provided.
Why It Matters: What Meta's Move Does to the Market
At the close of H1 2026, Meta's hiring campaign has produced three effects on the US AI talent market that extend beyond Meta's own headcount.
It reset the floor for all senior lateral comp. When Ruoming Pang moves from Apple to MSL for a reported $200 million, Apple has to revise what it offers the next Ruoming Pang it is trying to retain. When Shengjia Zhao leaves OpenAI, OpenAI has to ask what number would have kept him. The DeepMind.ai/the-batch reported in late 2025 that Meta's hiring spree had raised compensation across the board for top AI engineers and executives — not just within Meta. Anthropic's CEO Dario Amodei publicly stated in August 2025 (per Fortune) that the company was "not willing to compromise our compensation principles" in response to Meta's offers; xAI adjusted senior researcher packages upward in the same period [ENTRA estimate based on H-1B filing comparisons and Levels.fyi data, Q4 2025 vs Q2 2025].
It forced a structural test on non-financial retention levers. OpenAI, Anthropic, and Google DeepMind have each responded to Meta's checkbook with arguments that are harder to quantify: mission alignment, publication access, model quality, and team culture. Anthropic's offer to researchers emphasizes frontier safety research and the credibility of its constitutional AI work. DeepMind's response has been noncompete enforcement (six-to-twelve-month paid garden-leave clauses for departing senior researchers, per multiple 2025 reports) and retention bonuses [ENTRA estimate: upper range for principal researchers in the $500K–$900K band, based on aggregated Levels.fyi and H-1B data; no single named primary source has disclosed the exact ceiling] — plus a public commitment to maintaining publication norms. OpenAI's conversion from its Profit Participation Unit (PPU) equity structure toward standard RSUs — accelerated by the company's public benefit corporation transition in late 2025 — made its packages more legible to candidates comparing offers [per Levels.fyi PPU explainer and public reporting on OpenAI's PBC conversion; the full equity structure transition timeline extends into 2026].
The geography of the competition is specific. TBD Lab is based in Menlo Park, at Meta's main campus on Willow Road. Muse Spark was developed there. The New York office — located in Manhattan — handles applied research and product integration under Friedman's unit. For researchers based in Seattle or considering Microsoft Research as an alternative, Meta has moved less aggressively; the Seattle AI labor market is dominated by Microsoft and Amazon at the senior research level, and Meta has not opened an MSL node in Seattle. That gap is visible in the sourcing data: the notable MSL lateral hires in H1 2026 came from Menlo Park, San Francisco, and London-based DeepMind offices — not from Redmond.
Microsoft Research's position warrants a direct note. The organization has not been a primary source for MSL hires in H1 2026 — the main flows have been OpenAI, DeepMind, and frontier startups. That may reflect MSR's compensation structure (more conservative than the frontier lab bands) or the nature of the research agendas at play. Either way, Microsoft Research has lost fewer senior researchers to Meta than DeepMind or OpenAI in the period covered by this analysis.
What's Next: Three Signals for H2 2026
1. Whether TBD Lab expands beyond 50. Alexandr Wang's stated rationale for the October 2025 600-role reduction inside MSL was that smaller teams move faster. TBD Lab at 50 is a deliberate constraint. If Meta begins posting TBD Lab roles at scale in H2 2026, that signals the lab has shifted from assembly to production mode — and that a second wave of lateral hires from competitors is likely. Watch Meta's MSL job postings on LinkedIn and the Meta Careers page; the current postings (AI Research Scientist – Speech and Audio, Research Engineer Post-Training, Research Scientist AI Language) reflect the Muse Spark follow-on roadmap, not a new expansion.
2. The Avocado and Mango model releases. MSL's internal roadmap for H1 2026 named two models beyond Muse Spark: "Avocado," a reasoning-focused LLM targeting the logic and coding gap with GPT-5.4 and Claude Opus 4.6, and "Mango," a multimodal world model targeting generative video. Both were flagged for H1 2026 release but had not shipped as of this writing (June 19, 2026). Delays would raise questions about TBD Lab's research velocity relative to the pace of hiring and comp investment.
3. The Yann LeCun talent shadow. LeCun departed Meta in November 2025 (confirmed November 19, 2025, per CNBC), following the restructuring that placed FAIR under Wang. He subsequently co-founded Advanced Machine Intelligence Labs (AMI Labs) in Paris, which raised $1.03 billion in a March 2026 seed round at a $3.5 billion pre-money valuation (per TechCrunch, March 9, 2026). AMI Labs will recruit actively in H2 2026 — and LeCun's contact list is, by definition, a direct subset of the researchers Meta most wants to retain. The question is whether AMI can compete with TBD Lab's comp tier for researchers who are also weighing the non-financial argument of working with LeCun on architecture they find intellectually distinct from current scaling approaches.
Meta has, in nine months, built a research unit that competes directly with Anthropic and OpenAI for the same finite pool of senior frontier researchers — and it has done so by making the financial argument loud enough that every competitor has had to address it; whether the research argument is equally strong is what H2 2026 will answer.