Open research
A source-backed data hub for the companies, labs, benchmarks, and market signals pushing science toward autonomous discovery. Science-first AGI progress, tracked where it changes research workflows.
Capabilities & AGI progress
This section turns frontier-model progress into a compact AGI signal map: reasoning, agency, coding/R&D leverage, multimodal grounding, verification, and deployment readiness.
It separates a general cognitive AGI readiness estimate from science-first bottlenecks, where verification, reproducibility, and lab deployment carry a heavier penalty.
AGI progress signal
Public signal strength across AGI bottlenecks. The solid line is evidence through 2026; dashed lines are scenario overlays to 2032.
Current
62%
Uneven readiness
General cognitive AGI estimate
Cognitive and digital AGI: broad strong-human-level intellectual work, transfer across domains, tool use, multi-step planning, and enough reliability for real deployment. This is not superintelligence, full human replacement, or embodied robotics.
Readiness
63.4%
Rounded 63%; uncertainty 58-68%
Why 63%
Reasoning, coding, multimodal work, and economic knowledge tasks are already strong enough to make 55% too low.
Why not 70%+
Long-horizon agency, reliability, calibration, and robust real-world execution remain the binding constraints.
Science-first discount
A science-first AGI estimate falls closer to 53% because verification, reproducibility, scientific autonomy, and lab deployment dominate.
The geometric mean keeps weak bottlenecks visible: agency and reliability pull the composite down even when benchmark and coding scores are high.
100 * 0.72^0.18 * 0.71^0.18 * 0.76^0.14 * 0.70^0.10 * 0.55^0.18 * 0.48^0.16 * 0.55^0.06 = 63.4%General reasoning & knowledge
Strong GPQA, AIME, and ARC-AGI-1 progress; ARC-AGI-2 and HLE still leave headroom.
w 18%
72/100
Economic knowledge work
GDPval shows frontier models approaching expert work products across many occupations.
w 18%
71/100
Coding & tool use
SWE-bench Verified and related tool-use evals are among the most mature public signals.
w 14%
76/100
Multimodal/context handling
Vision, long-context, document, and screen workflows are much stronger, but not yet universal world-modeling.
w 10%
70/100
Long-horizon agency
The main bottleneck: autonomous tasks are still short, scaffolded, or well-specified.
w 18%
55/100
Reliability/calibration/safety
Hallucination, overconfidence, brittle behavior, and hard-to-detect errors still limit delegation.
w 16%
48/100
Deployment / real-world integration
Productivity is already widespread, but not full autonomous replacement of roles.
w 6%
55/100
Percent is curated public signal strength and editorial readiness, not measured AGI attainment or probability.
Apr 2026
General model intelligence
Composite score across agents, coding, science, reasoning, knowledge, and instruction following
Provides a production-oriented view of which frontier models are strong enough to act as reasoning engines inside scientific agents.
Current signal
Live leaderboard score, provider, price, speed, latency, and context-window tracking
Apr 2026
Capability trends
Benchmark results across 40+ evaluations, with internal and external result provenance
Useful for seeing whether scientific reasoning, agentic work, math, coding, and multimodal capabilities are improving fast enough to change lab workflows.
Mar 2026
Mathematics
Accuracy on extremely difficult math problems and open-problem variants
High-end mathematical reasoning is one of the cleanest proxies for whether models can contribute to formal scientific discovery.
Apr 2026
Scientific reasoning
Expert-level graduate science multiple-choice accuracy
Directly probes PhD-level physics, chemistry, and biology reasoning, though it remains a static question-answer benchmark.
Mar 2026
Scientific coding
Pass rate on scientific programming tasks
Measures whether models can turn scientific specifications into executable code, a core dependency for autonomous analysis and simulation.
Apr 2026
Agentic software engineering
Resolved real GitHub issues
Software-engineering agents are a leading indicator for whether models can operate long-horizon scientific toolchains and repair failed experiments.
Mar 2026
Long-horizon agents
Pass@1 task completion in realistic multi-application workflows
Lab automation requires agents that coordinate files, tools, state, and multi-step objectives rather than answering isolated prompts.
Mar 2026
Cross-domain expert reasoning
Accuracy on hard expert-written questions
A broad stress test for frontier models, useful only when interpreted alongside domain-specific science benchmarks and tool-use evaluations.
Market snapshot
The field is no longer a loose list of AI labs and automation vendors. It is splitting into a stack: reasoning engines, experimental execution layers, verification systems, proprietary data loops, and institutional infrastructure.
This release keeps the editorial map, but adds comparable dimensions and source-backed records so the page can evolve into a benchmark institute rather than a static directory.
Entities
26
Company funding
14
Ecosystem events
6
Milestones
115
Sources
40
Includes closed company rounds plus ecosystem capital events with explicit committed amounts. Up-to and valuation-only events stay out of totals.
A single entity can occupy multiple roles. Full-stack labs count across reason, execute, verify, and data where the public record supports it.
Region
Stage
Scores are dimensions, not a ranking. They make tradeoffs legible before a future Scivity index exists.
Autonomy stack
The useful comparison is not “AI company versus lab automation company.” It is which part of the scientific method each system can own, and which parts still depend on people or external infrastructure.
Agents and models that generate hypotheses, plans, code, protocols, papers, or scientific arguments.
Physical and remote systems that turn plans into experiments, synthesis runs, assays, or lab workflows.
Evaluation, replication, critique, measurement, and validation layers that determine whether a claim survives contact with reality.
Platforms that generate, structure, or own proprietary scientific data loops unavailable on the open web.
Cloud labs, national facilities, robotics stacks, orchestration software, and institutional programs that make autonomous science deployable.
Entity explorer
Filter by stack role, domain, organization type, evidence level, capital band, and access model. Select up to four entities for a side-by-side comparison.
01
Public company
Public · North America · Boston, MA · Founded 2008
Synthetic biology foundry. Engineered organisms as a service; increasingly AI-augmented across the design-build-test-learn cycle.
Autonomy
3 / 4Execution
4 / 4Verification
3 / 4Scientific evidence
4 / 4Capital
3 / 4Traction
4 / 402
Company
Private scaleup · North America · San Francisco, CA · Founded 2025
Autonomous physical labs for materials and chemistry. AI scientists generate hypotheses; robotic labs run synthesis and characterization; results feed back as training data unavailable anywhere else.
Autonomy
4 / 4Execution
4 / 4Verification
3 / 4Scientific evidence
3 / 4Capital
4 / 4Traction
2 / 403
Company
Private scaleup · North America · Cambridge, MA · Founded 2023
AI Science Factories — integrated reasoning models, robotic labs, and verifiers running the scientific method across life, chemistry, and materials sciences.
Autonomy
4 / 4Execution
4 / 4Verification
3 / 4Scientific evidence
3 / 4Capital
4 / 4Traction
2 / 404
Company
Growth · Europe · Glasgow, UK · Founded 2023
Chemputer platform — universal chemical synthesis via programmable hardware and AI-driven reaction planning. Commercialization of Lee Cronin's Glasgow work.
Autonomy
3 / 4Execution
4 / 4Verification
3 / 4Scientific evidence
4 / 4Capital
3 / 4Traction
3 / 405
Public company
Public · North America · Salt Lake City, UT · Founded 2013
Publicly traded AI-native drug discovery platform. Industrial-scale phenomics, ML-driven target identification, and image-based screening at PB-scale.
Autonomy
2 / 4Execution
3 / 4Verification
3 / 4Scientific evidence
4 / 4Capital
4 / 4Traction
4 / 406
Government lab
Research facility · North America · Berkeley, CA · Founded 2022
Autonomous materials synthesis lab at Lawrence Berkeley National Lab. AI-driven synthesis of novel inorganic materials.
Autonomy
4 / 4Execution
4 / 4Verification
2 / 4Scientific evidence
4 / 4Capital
2 / 4Traction
3 / 407
Frontier AI lab
Big Tech lab · Europe · London, UK · Founded 2010
AlphaFold, AlphaProteo, GNoME, AlphaGeometry, AlphaEvolve — state-of-the-art scientific AI, without a dedicated autonomous-lab product line.
Autonomy
3 / 4Execution
1 / 4Verification
3 / 4Scientific evidence
4 / 4Capital
4 / 4Traction
4 / 408
Public company
Public · Asia-Pacific · Hong Kong (also Boston, New York) · Founded 2014
End-to-end generative AI platform for drug discovery and development (Pharma.AI). Public on the Hong Kong Stock Exchange since December 2025. TNIK inhibitor rentosertib is the first AI-discovered and AI-designed drug to deliver Phase 2a clinical data.
Autonomy
2 / 4Execution
2 / 4Verification
3 / 4Scientific evidence
4 / 4Capital
4 / 4Traction
4 / 409
Company
Commercial · North America · South San Francisco, CA · Founded 2010
Remote-controlled cloud lab — programmable chemistry and biology infrastructure accessed via a full scientific-computing environment.
Autonomy
2 / 4Execution
4 / 4Verification
3 / 4Scientific evidence
3 / 4Capital
2 / 4Traction
4 / 410
Academic network
Research network · North America · Toronto, Canada · Founded 2021
International consortium based at University of Toronto. Coordinates self-driving labs across member institutions; publishes protocols and standards.
Autonomy
2 / 4Execution
3 / 4Verification
3 / 4Scientific evidence
4 / 4Capital
3 / 4Traction
3 / 411
Frontier AI lab
Frontier platform · North America · San Francisco, CA · Founded 2015
Frontier model developer. April 2026 Industrial Policy paper explicitly calls for distributed AI-enabled laboratories as public infrastructure.
Autonomy
3 / 4Execution
1 / 4Verification
3 / 4Scientific evidence
3 / 4Capital
4 / 4Traction
4 / 412
Company
Private scaleup · Europe · London, UK · Founded 2021
Alphabet-backed AI drug discovery company spun out of Google DeepMind. Builds on AlphaFold to design novel therapeutics, with a 17-program pipeline advancing toward the clinic.
Autonomy
2 / 4Execution
2 / 4Verification
3 / 4Scientific evidence
4 / 4Capital
4 / 4Traction
3 / 413
Company
Private scaleup · Asia-Pacific · Tokyo, Japan · Founded 2023
The AI Scientist — autonomous ML research agent. Produces end-to-end research papers with minimal human involvement.
Autonomy
4 / 4Execution
1 / 4Verification
2 / 4Scientific evidence
3 / 4Capital
4 / 4Traction
3 / 414
Nonprofit lab
Research platform · North America · San Francisco, CA · Founded 2023
Nonprofit AI research lab building autonomous scientists for biology. Released the Crow, Falcon, Owl, and Phoenix agent family. Commercial spinout Edison Scientific in late 2025.
Autonomy
4 / 4Execution
1 / 4Verification
3 / 4Scientific evidence
4 / 4Capital
2 / 4Traction
3 / 415
Government lab
Research facility · North America · Richland, WA · Founded 1965
Pacific Northwest National Laboratory. DOE lab with autonomous experimentation programs in chemistry and energy. Part of the DOE × DeepMind Genesis Mission network.
Autonomy
2 / 4Execution
3 / 4Verification
3 / 4Scientific evidence
4 / 4Capital
2 / 4Traction
3 / 416
Company
Growth · Asia-Pacific · Singapore · Founded 2022
AI-for-science startup running a 24/7 autonomous chemistry synthesis line. Serves as the chemistry discovery engine for pharma; global HQ and flagship self-driving lab in Singapore.
Autonomy
3 / 4Execution
4 / 4Verification
2 / 4Scientific evidence
2 / 4Capital
3 / 4Traction
3 / 417
Company
Growth · Europe · London, UK · Founded 2015
Lab-automation operating system for AI-ready wet labs. Modular robotics, orchestration software, and unified data infrastructure to turn physical labs into programmable, repeatable systems.
Autonomy
2 / 4Execution
4 / 4Verification
2 / 4Scientific evidence
2 / 4Capital
2 / 4Traction
4 / 418
Company
Private scaleup · North America · South San Francisco, CA · Founded 2024
AI-driven drug discovery and design company. Uses ML from hypothesis through clinical development.
Autonomy
2 / 4Execution
2 / 4Verification
3 / 4Scientific evidence
3 / 4Capital
4 / 4Traction
2 / 419
Company
Seed · North America · San Francisco, CA · Founded 2025
Commercial spinout of FutureHouse. Kosmos is a multi-agent AI co-scientist that runs 12-hour cycles of literature review, data analysis, and hypothesis generation — executing ~42,000 lines of code and reading ~1,500 papers per run.
Autonomy
4 / 4Execution
1 / 4Verification
3 / 4Scientific evidence
3 / 4Capital
3 / 4Traction
2 / 420
Company
Commercial · North America · San Diego, CA · Founded 2012
Robotic cloud lab for life sciences. Automated workflows for drug discovery, synthesis, and assays, accessible via API.
Autonomy
2 / 4Execution
4 / 4Verification
2 / 4Scientific evidence
2 / 4Capital
2 / 4Traction
3 / 421
Company
Seed · North America · San Francisco, CA · Founded 2023
Autonomous agents for ML research. Carl ideates and writes papers; Mira embeds new research directly into production ML models.
Autonomy
4 / 4Execution
1 / 4Verification
2 / 4Scientific evidence
3 / 4Capital
2 / 4Traction
2 / 422
Company
Commercial · Europe · Lausanne, Switzerland · Founded 2019
Self-driving lab orchestration software. Bayesian optimization and closed-loop experimentation for chemistry, materials, and pharma R&D teams.
Autonomy
3 / 4Execution
2 / 4Verification
3 / 4Scientific evidence
3 / 4Capital
1 / 4Traction
2 / 423
Company
Seed · North America · New York, NY · Founded 2022
AI platform for materials discovery. Focus on industrial biomanufacturing and novel functional materials.
Autonomy
3 / 4Execution
3 / 4Verification
2 / 4Scientific evidence
2 / 4Capital
2 / 4Traction
2 / 424
Company
Seed · North America · San Francisco, CA · Founded 2025
Applied research lab automating parts of the ML research workflow — experiment planning, training, evaluation, iteration.
Autonomy
3 / 4Execution
1 / 4Verification
2 / 4Scientific evidence
2 / 4Capital
1 / 4Traction
1 / 425
Frontier AI lab
Seed · Europe · Paris, France · Founded 2025
Advanced Machine Intelligence — Yann LeCun's new lab. Building world models based on Joint Embedding Predictive Architecture (JEPA) rather than next-token LLMs; aims to learn from physical reality as the path to human-level intelligence.
Autonomy
2 / 4Execution
1 / 4Verification
1 / 4Scientific evidence
1 / 4Capital
4 / 4Traction
1 / 426
Company
Seed · North America · San Francisco, CA · Founded 2025
Natural-language-to-robot-arm layer for lab automation. Scientists type protocols in English; software programs off-the-shelf robotics to execute them.
Autonomy
2 / 4Execution
3 / 4Verification
1 / 4Scientific evidence
1 / 4Capital
1 / 4Traction
1 / 4Timeline
The timeline combines company milestones, field-wide events, and ecosystem capital signals. Cross-market events stay separate from company funding so charts and foreign keys remain clean.
Formal correction (650(8100):E1). Paper no longer claims materials necessarily novel to science. The field's defining verification failure is officially recorded.
AI co-scientist deployed across all 17 US National Labs via Gemini for Government. Stated goal: double US scientific productivity within a decade.
Edison released Kosmos (arXiv 2511.02824) — a multi-agent AI scientist that executes ~42,000 lines of code and reads ~1,500 papers per run. Launch showcased seven discoveries across neuroscience, materials, and statistical genetics: three reproducing human-led findings and four entirely novel.
Demis Hassabis and John Jumper shared the 2024 Nobel Prize in Chemistry (jointly with David Baker) for AlphaFold's protein structure prediction — the first Nobel to recognize an AI-driven scientific breakthrough.
Apr 2026
12 projects pairing AI with self-driving labs to compress catalyst-development timelines from ~10 years to ~1 year. Recipients include NC State ($2.99M), UW-Madison ($2.84M), Ames National Lab ($2.52M).
Apr 2026
All-stock. First Anthropic M&A into a new domain. ~10 employees, ~$40M+ per employee. Stanton + Frey (ex-Prescient Design). Dimension VC reportedly 38,513% IRR.
Apr 2026
All-stock, ~$400M. First Anthropic M&A into a new domain. Signals that frontier labs will enter AI-for-science domain-by-domain through acqui-hires.
Apr 2026
Calls for 'distributed AI-enabled laboratories' as public infrastructure. Commits $100K grants + $1M API credits, not capital to build the network.
Apr 2026
First large federal bet explicitly tying autonomous labs to industrial catalyst development. Targets compressing catalyst discovery from ~10 years to ~1 year across universities and national labs.
Apr 2026
First commercial product from Sakana. Autonomous enterprise research agent (Adaptive Branching MCTS + AI Scientist workflow) producing slide decks and multi-dozen-page reports in one 8-hour run — the AI Scientist architecture repurposed for corporate strategy.
Apr 2026
Cronin/Chemify published three peer-reviewed papers in PNAS and Nature Communications sister journals validating the chemputer for autonomous drug synthesis and kinase inhibitor discovery.
Apr 2026
GPT-Rosalind, a specialized biology reasoning model scoring 0.751 on BixBench (ahead of GPT-5.4), launched with Amgen, Moderna, Thermo Fisher, and the Allen Institute as partners.
Apr 2026
ProQR announced an AI-enabled drug discovery partnership giving it access to Ginkgo's Nebula autonomous lab to generate high-throughput data for its Axiomer RNA editing platform. Ginkgo also took a strategic equity stake. ProQR concurrently formed an AI Advisory Board.
Mar 2026
Toyota Ventures, Perplexity Fund, S32, MaC participated. Commercial launch of Carl and Mira agents.
Mar 2026
Yann LeCun's new lab. $3.5B pre-money — the largest seed round in European tech history. Signal that frontier-lab veterans are entering the science-AI space outside the Big Tech umbrella.
Mar 2026
In talks at ~$7B — a 5.4× jump from the $1.3B seed closed six months earlier. Not yet closed; reported by Bloomberg.
Mar 2026
$115M upfront plus milestones. 28+ pipeline drugs.
Mar 2026
DE-FOA-0003612. Phase I ($500K–$750K, 9 months) and Phase II ($6M–$15M, 3 years) awards across 20+ national challenges in manufacturing, biotech, critical materials, nuclear, and quantum. Operationalizes the Genesis Mission announced in late 2025.
Mar 2026
Nature 651:914–919. A human-authored description of the AI Scientist system, commonly misreported as an AI-authored Nature paper.
Mar 2026
PNNL publicly detailed the Autonomy Studio — a purpose-built LDRD-funded facility for design-build-test-learn cycles with digital twins, AI agents, and robotic platforms across biological engineering and critical minerals.
Mar 2026
Xaira released X-Cell, a 4.9B-parameter diffusion-based virtual cell model trained on X-Atlas/Pisces — the largest-ever CRISPRi Perturb-seq dataset at 25.6M single-cell transcriptomes.
Mar 2026
Yann LeCun's Paris-based lab launched publicly with a $1.03B seed at $3.5B pre-money — the largest seed in European tech history. Architecture bet is explicitly against the LLM paradigm, building on Joint Embedding Predictive Architecture (JEPA). Research bench includes Saining Xie, Pascale Fung, and Michael Rabbat.
Mar 2026
Ginkgo opened web-browser access to Nebula, its 50+ instrument autonomous lab in Boston, via Ginkgo Cloud Lab. Protocols are submitted in natural language and assessed for fleet compatibility by an AI agent called EstiMate. Ginkgo plans to move all R&D services onto Nebula and decommission traditional benches.
Mar 2026
DOE published FOA DE-FOA-0003612 for the Genesis Mission, inviting Phase I and Phase II proposals across 20+ AI-for-science national challenges. Phase I awards up to $750K for nine-month projects; Phase II up to $15M over three years. Applications due April 28, 2026.
Mar 2026
A Nature Technology Feature (30 March 2026) surveys the maturing self-driving-lab industry — profiling King's Eve at Chalmers, CMU's Coscientist on Emerald Cloud Lab, and the commercial wave behind them — while acknowledging the limits that still require human scientists.
Feb 2026
Formal correction (650(8100):E1). Paper no longer claims materials necessarily novel to science. The field's defining verification failure is officially recorded.
Feb 2026
Danaher subsidiary Beckman Coulter Life Sciences integrated its liquid handling and genomic instruments into Automata's Linq platform, announced at SLAS 2026.
Feb 2026
Ginkgo's autonomous lab driven by GPT-5 ran 36,000 cell-free protein synthesis experiments and achieved a 40% reduction in sfGFP production cost, beating state-of-the-art benchmarks.
Feb 2026
In its Q4 2025 report, Ginkgo announced a strategic pivot to make autonomous labs its core offering and plans to divest biosecurity to concentrate capital on the platform.
Feb 2026
Isomorphic Labs announced IsoDDE, a unified drug-design platform reported to more than double AlphaFold 3's accuracy on protein-ligand structure prediction benchmarks.
Feb 2026
Isomorphic announced IsoDDE — an integrated drug design engine claimed to more than double AlphaFold 3 on protein-ligand binding prediction and to cover the full discovery workflow end-to-end. Pipeline of 17 programs now spans oncology, immunology, and cardiovascular disease.
Jan 2026
Danaher Ventures, Tru Arrow Partners, Octopus Ventures, Entrepreneurs First participated. Lab-automation OS for AI-ready wet labs; five top pharma customers. Danaher joins the board.
Jan 2026
AI-driven drug discovery partnership.
Jan 2026
Co-innovation lab partnership. Picks-and-shovels strategy across pharma.
Jan 2026
Radical AI established New York's first fully autonomous materials science laboratory at the Brooklyn Navy Yard, with capacity for ~100 AI-driven experiments per day.
Jan 2026
Prism, a free LaTeX-native collaborative research workspace powered by GPT-5.2, targets drafting, revising, and publishing scientific papers.
Jan 2026
Autoscience announced Mira, an automated ML research scientist designed to invent new model architectures. Mira was entered into a live $50k Kaggle competition and climbed to a Bronze Medal autonomously.
Dec 2025
FutureHouse commercial spinout. $250M valuation. Commercializes the Kosmos AI co-scientist. Angels: Jeff Dean, Dmitri Alperovitch.
Dec 2025
Raise paired with global HQ move and new self-driving lab in Singapore.
Dec 2025
Meta acquired Butterfly Effect (Manus) around Dec 29–30, 2025. Meta to discontinue China operations.
Methodology
Included entities must materially affect autonomous scientific discovery: reasoning systems, wet-lab or dry-lab execution layers, scientific verification, domain data loops, or infrastructure that makes autonomous research deployable.
Pure AI infrastructure providers, broad model labs, and general AGI benchmarks are included only when they have a clear science-relevant signal. The aim is a science-first AGI-progress lens, not a general model directory.
Autonomy
How much of the scientific workflow can run without human step-by-step operation.
Execution
Whether the system can touch the physical world, run labs, trigger assays, or operate production scientific workflows.
Verification
How directly the system can evaluate, reproduce, critique, or measure scientific claims.
Scientific evidence
Strength of public evidence: papers, clinical data, benchmark results, open artifacts, or deployed research outputs.
Capital
Scale of committed capital, strategic partnerships, public-market access, or institutional funding.
Traction
Commercial, institutional, research, or user adoption signals beyond a launch announcement.
Every entity, funding event, milestone, ecosystem event, and benchmark record carries source IDs. The validator checks provenance links, date shape, foreign keys, and score ranges.
v1 is manually curated. Codex automations and recurring refresh flows are intentionally deferred until the data contract is stable.
Funding totals exclude valuation-only discussions and milestone-contingent “up to” amounts. Cross-market events are stored separately from company funding to keep company foreign keys clean.
This page is free to reuse under Creative Commons BY 4.0. Citation: Scivity Labs (2026), “The AI for Science Landscape,” scivity.org/landscape.
40 source records are currently tracked.