Skip to content
← AI by field
Biologists, drug discovery teams and biotech

AI for biology and drug discovery, right now

Designing proteins and molecules, reading genomes, planning experiments and running lab work.

No benchmark of its own yetUpdated Nov 12, 2026 from 9 sources
Open biology in Spezia Free account. No card.
Where AI is in biology now

AI models now design proteins and predict variant effects, but no AI-designed drug is FDA-approved.

What models can do now

  • Google DeepMind's AlphaGenome 2 predicts variant effects across hundreds of human cell types and adapts to unseen cell and disease states without retraining.[1][9]
  • Machine learning tools designed active cysteine proteases and metalloproteases from scratch.[3]
  • An AI-guided AutoDNA lab completed a DNA read-write cycle in about 200 hours with expert-level precision.[2]
  • Researchers used AlphaFold-Multimer to predict 2,895 likely interactions among human mitochondrial proteins.[5]
  • Anthropic says Claude helped determine a 3.2-angstrom vault particle structure from a bacteriophage protein.[7]
  • Google DeepMind released Antigravity drug discovery tools that let an AI agent propose molecules in about an hour.[8]

Where they still fall short

  • Big Pharma CEOs say AI cannot yet predict human biology or safety, and no AI-designed drug is FDA-approved.[4]
  • Pharma executives say AI drug discovery falls short, citing Vertex test of 50 AI companies with no notable results.[6]

What is expected next

  • Google DeepMind expects AlphaGenome 2 to help researchers pinpoint previously overlooked disease-relevant variants with high confidence.[9]

How fast it is moving

The record on this field's benchmarks over the last year, from Spezia's model leaderboard.

Few public benchmarks test biology research directly. These are the closest: regulatory work, scientific coding and graduate-level science questions.

LifeSci Diamond

55 regulatory affairs tasks from drug and device makers, such as FDA and EMA submissions, labeling and compliance, with answer keys written or reviewed by practising specialists.

Record: 20%, up from 0% 6 months ago (GPT-6 Astra)

0%5%10%15%20%GPT-6 AstraGPT-5.6 Sol
Apr 26Jul 26Oct 26

the record so far set a record other models

SciCode

Research coding problems written by scientists across 16 fields, from physics and chemistry to biology and materials science.

Record: 91.8%, up from 44.8% a year ago (2×) (Mistral Large 4)

0%25%50%75%100%Mistral Large 4Seed 2.1 Pro
Oct 25Jan 26Apr 26Jul 26Oct 26

the record so far set a record other models

  • GPQA Diamond96.5%, up from 87.6% a year ago · Gemini 4 Carbon

The best models for biology now

On the closest general benchmarks, since none measures this field directly.

  1. 1GPT-6 Astra
  2. 2GPT-6.1 Sol
  3. 3GPT-6.1 Terra
  4. 4Gemini 4 Argon
  5. 5Gemini 3.7 Flash
  6. 6Grok 4.6

Milestones

What became possible, newest first, each marked by how firm it is. Open one for the words it rests on.

Showndemonstrated with published results, a paper or an independent checkClaimedsaid by the company or people who built it, not yet checked by othersReportedreported by others, not confirmed by the company or a published resultNot publicdone by a model that is not released to the public

  1. Nov 11, 2026ClaimedAlphaGenome 2
    Google DeepMind announced AlphaGenome 2, a DNA sequence model predicting variant effects in specific cell types.
    “Google DeepMind announces AlphaGenome 2, a DNA sequence model predicting variant effects in specific cell types.” Google DeepMind announces AlphaGenome 2, a DNA sequence model predicting variant
  2. Nov 10, 2026Reported
    AI-guided AutoDNA lab completed a DNA read-write cycle in about 200 hours with expert-level precision.
    “AI-guided AutoDNA lab completed a DNA read-write cycle in about 200 hours with expert-level precision.” AI-guided AutoDNA lab completed a DNA read-write cycle in about 200 hours with e
  3. Nov 9, 2026Reported
    Machine learning tools used to design active cysteine proteases and metalloproteases from scratch.
    “Machine learning tools used to design active cysteine proteases and metalloproteases from scratch.” Machine learning tools used to design active cysteine proteases and metalloprote
  4. Nov 8, 2026Reported
    Multimodal AI-guided enzyme engineering boosts CO2 biocatalysis into l-malate at 22.7 g/L titer.
    “Multimodal AI-guided enzyme engineering boosts CO2 biocatalysis into l-malate at 22.7 g/L titer.” Multimodal AI-guided enzyme engineering boosts CO2 biocatalysis into l-malate at
  5. Nov 8, 2026ReportedAlphaFold-Multimer
    Researchers used AlphaFold-Multimer to predict 2,895 likely interactions among human mitochondrial proteins.
    “Researchers used AlphaFold-Multimer to predict 2,895 likely interactions among human mitochondrial proteins.” Researchers used AlphaFold-Multimer to predict 2,895 likely interactions among h
  6. Nov 8, 2026ReportedWorldParticle
    MIT and UBC researchers unveil WorldParticle, one AI model simulating sand, fluids and proteins.
    “a single deep-learning system, called WorldParticle, that could help engineers model different physical phenomena, from sliding sand to folding proteins” MIT and UBC researchers unveil WorldParticle, one AI model simulating sand, flui
Every milestone, story and model Pick a model to see what it can do in biology.

Sources

  1. [1]Google DeepMind announces AlphaGenome 2, a DNA sequence model predicting variant effects in specific cell types.Spezia news · 2026-11-11
  2. [2]AI-guided AutoDNA lab completed a DNA read-write cycle in about 200 hours with expert-level precision.Spezia news · 2026-11-10
  3. [3]Machine learning tools used to design active cysteine proteases and metalloproteases from scratch.Spezia news · 2026-11-09
  4. [4]Big Pharma CEOs say AI cannot yet predict human biology or safety, and no AI-designed drug is FDA-approved.Spezia news · 2026-11-08
  5. [5]Researchers used AlphaFold-Multimer to predict 2,895 likely interactions among human mitochondrial proteins.Spezia news · 2026-11-08
  6. [6]Pharma executives say AI drug discovery falls short, citing Vertex test of 50 AI companies with no notable results.Spezia news · 2026-11-08
  7. [7]Anthropic says Claude helped determine a 3.2-angstrom vault particle structure from a bacteriophage protein.Spezia news · 2026-11-06
  8. [8]Google DeepMind released Antigravity drug discovery tools that let an AI agent propose molecules in about an hour.Spezia news · 2026-11-05
  9. [9]@pushmeet on XX · 2026-11-11

Other fields