Quarterly editorial briefings on the AI labs at the frontier -- how they hire, what they pay, what they're shipping, and where the engineering talent is moving. Written by the OpenTalent network from the inside.
// Latest issue -- Issue 14
What changed inside frontier labs this quarter -- and what it means for the engineers building the next wave.
Anthropic, OpenAI, Google DeepMind, Mistral, and three frontier startups quietly shifted what they're willing to pay for. We mapped 2,400 senior AI engineering hires from January to April. Here's where compensation rose, where it fell, and which specializations the labs are now competing for hardest.
Total comp ranges across pre-training, post-training, agents, RL, and applied AI roles -- by lab tier and geography.
// Labs we track
Hiring velocity, talent flow, research output, and stated direction -- sourced from public signals and from the engineers inside the OpenTalent network.
Safety-first frontier lab; product velocity has shifted toward agents and Claude Code in the last two quarters.
Aggressive expansion into applied AI; reasoning-model team is now the largest single research org.
Pre-training scale push continues; renewed agentic-Gemini hiring across London and Mountain View.
European frontier lab, post-training and openness focus; selective senior hiring at the Paris HQ.
Compute-heavy pre-training bet; recruiting at the senior staff and principal levels with above-market offers.
Enterprise-focused frontier shop; deep hiring in retrieval, evals, and applied agent infrastructure.
Aggressive applied-AI growth; recruiting senior RAG and product engineers across SF and Bangalore.
Reorganized post-acquisition; consumer AI scale-up under MS AI banner.
Multimodal frontier lab; expanding video and embodied-AI research at the Singapore HQ.
India's frontier lab; deep hiring in Indic languages, evals, and post-training for Bharat-scale deployments.
Generative-video frontier; senior research and infra hires across NYC, London, and remote.
Data layer for the frontier; renewed emphasis on RLHF/RLAIF datasets and verifier engineering.
// Compensation benchmarks
Total compensation ranges (base + equity + bonus, annualized) across major frontier labs, for engineers with 5-8 years of experience. Sourced from 2,400+ network-verified offers and conversations.
Numbers reflect senior IC offers in US-based roles. Geo-adjustments and senior staff / principal levels are covered in the linked briefings.
| Role | Range (USD, total) | Median | YoY |
|---|---|---|---|
| Pre-training research engineer | $650K -- $1.2M | $880K | +14% |
| Post-training / RLHF engineer | $520K -- $950K | $720K | +11% |
| Agent / tool-use engineer | $480K -- $850K | $660K | +19% |
| RL environments / verifier engineer | $500K -- $880K | $690K | +22% |
| ML infrastructure engineer | $460K -- $820K | $640K | +8% |
| Multimodal / VLA engineer | $470K -- $830K | $650K | +13% |
| Applied AI engineer (RAG, evals, prod) | $380K -- $680K | $520K | +9% |
// Trend signals
From the May 2026 dataset. We track these every quarter; here are the four worth knowing.
YoY growth in median comp for RL environments / verifier engineers.
Of senior offers we tracked included a signing bonus over $200K -- up from 21% in 2025.
Increase in cross-lab senior engineering moves vs. same window last year.
Median frontier-lab senior-IC offer cycle, scope to written offer. Was 27 days in 2024.
// All briefings
Filter by topic above. New briefings published roughly every two weeks.
Mapping 2,400 senior AI hires from January to April. The roles, the labs, the compensation shifts, and what it tells engineers planning their next move.
Total compensation ranges across pre-training, post-training, agents, RL, and applied AI roles -- by lab tier and geography.
Why the frontier labs quietly stopped reaching for off-the-shelf agent frameworks, and what they're building in-house instead. Implications for hiring.
Why frontier post-training teams are paying premium for verifier and reward-design engineers -- and what the skill stack actually looks like in 2026.
What the major labs are running on in 2026 -- Ray vs. Kueue vs. proprietary schedulers, FSDP vs. 3D parallelism, and which infra skills are now table-stakes for senior IC roles.
280+ senior cross-lab moves in the last quarter, mapped. The patterns, the magnets, and what the flow tells you about each lab's direction.
Synthetic data, MoE routing, and the architectures actually being trained inside the frontier labs this year.
Frontier safety teams are paying senior-staff compensation for adversarial-eval and interpretability engineers. What the work looks like -- and how to break in.
Paris, London, and Zurich have closed roughly 60% of the historical pay gap. The labs leading the change -- and what it means for hiring strategies.
Three major labs are pitching versions of an "agent OS" internally. We compare the stacks -- and the engineering skills each requires.
Speculative decoding, paged attention, and the inference-optimization skill stack the labs are now paying premium for.
Sarvam, Krutrim, and three quieter shops. The hiring picture, comp, and the realistic talent flow from US frontier labs back to India in 2026.
The engineers we profile here are the ones in our network. Hire them, or join them -- both start at OpenTalent.