Skip to content
← AI by field
Scientists and R&D teams

AI for physics, chemistry and materials, right now

Running simulations, writing research code, discovering materials and reasoning through hard science problems.

Moving: very fastUpdated Nov 12, 2026 from 10 sources
Open science in Spezia Free account. No card.
Where AI is in science now

AI models now solve graduate-level science problems and assist in real experiments, but still struggle with deep chemistry reasoning.

What models can do now

  • Claude Fable 5.2 leads science benchmarks, followed by Gemini 3.1 Pro and Claude Opus 5.[1]
  • NVIDIA's GR00T 2 controls dexterous hands to perform a chemistry lab experiment in real time.[2]
  • Anthropic says Claude helped solve two open physics problems involving black holes and quantum gravity.[3]
  • Google DeepMind and Quantum AI researchers used AI agents and Lean proofs to cut quantum program resource costs.[4]
  • An unreleased Anthropic model proved the 20-year-old Aaronson–Ambainis quantum conjecture, verified by Lean.[6]
  • Tokyo Tech AI agents planned and ran quantum spin experiments, deriving a calibration law and confirming it in repeat measurements.[7]

Where they still fall short

  • Top LLMs score poorly on chemistry reasoning, with Gemini 3 Pro achieving only 27.8% scope and 10.7% depth on ChemInsight.[5]
  • Claude Opus 4.5 still exhibits 'sloppy' behavior and requires expert supervision, so true end-to-end scientific autonomy is not yet achieved.[10]

What is expected next

  • A Princeton workshop aims to identify bottlenecks in AI-driven superconductor discovery and define a shared roadmap.[9]
  • China forum experts stress human oversight in scientific discovery as autonomous research agents advance.[8]

How fast it is moving

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

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 Diamond

Graduate-level questions in biology, physics and chemistry, written by PhDs so that searching the web does not give the answer away.

Record: 96.5%, up from 87.6% a year ago (Gemini 4 Carbon)

0%25%50%75%100%Gemini 4 CarbonGPT-5.4 Pro
Jul 25Oct 25Jan 26Apr 26Jul 26Oct 26

the record so far set a record other models

The best models for science now

On this field's benchmarks alone.

  1. 1Claude Fable 5.2
  2. 2Gemini 3.1 Pro
  3. 3Claude Opus 5
  4. 4GPT-5.5 Pro
  5. 5GPT-5.5
  6. 6Kimi K3.1 Pro

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 10, 2026ClaimedUnreleased Anthropic model
    An unreleased Anthropic model proved the 20-year-old Aaronson–Ambainis quantum conjecture, verified by Lean.
    “with the help of an unreleased Anthropic model, it proved the Aaronson–Ambainis conjecture, which has puzzled the mathematics community for nearly 20 years” An unreleased Anthropic model proved the 20-year-old Aaronson–Ambainis quantum c
  2. Nov 10, 2026ShownGR00T 2
    NVIDIA's GR00T 2 controlled dexterous hands to perform a chemistry lab experiment at real-time speed.
    “This is GR00T 2 controlling real dexterous hands to perform a chemistry lab experiment (and clean up after itself) end to end.” NVIDIA's Jim Fan shows GR00T 2 controlling dexterous hands through a chemistry l
  3. Nov 10, 2026ClaimedClaude
    Anthropic says Claude helped solve two open physics problems involving black holes and quantum gravity.
    “Anthropic says Claude helped them solve two open problems involving black holes and quantum gravity” Anthropic says Claude helped solve two open physics problems and formalize proof
  4. Nov 9, 2026ShownAI agents with Lean proofs
    Google DeepMind and Quantum AI researchers used AI agents and Lean proofs to cut quantum program resource costs.
    “Google researchers used artificial intelligence and computer-checked mathematical proofs to design quantum programs that require substantially fewer computing resources” Google DeepMind and Quantum AI researchers used AI agents and Lean proofs to cut
  5. Nov 8, 2026ShownAGH-Net
    An attention-gated NLP model detected dementia from speech with 80.18% accuracy and 83.87% sensitivity.
    “The proposed AGH-Net achieved 80.18% accuracy and a high clinical sensitivity of 83.87%.” Attention-gated NLP model detects dementia from speech with 80.18% accuracy and
  6. Nov 6, 2026ShownPhotonic quantum training head
    TuringQ and Shanghai Jiao Tong University used a photonic quantum training head to design working defensins and xylanases.
    “reports antifungal activity for three designed plant defensins and measured enzyme activity for designed xylanases and short-chain dehydrogenases” TuringQ and Shanghai Jiao Tong University used a photonic quantum training head
Every milestone, story and model Pick a model to see what it can do in science.

Sources

  1. [1]Best models on science benchmarksLeaderboard · 2026-11-12
  2. [2]NVIDIA's Jim Fan shows GR00T 2 controlling dexterous hands through a chemistry lab experiment at real-time speed.Spezia news · 2026-11-11
  3. [3]Anthropic says Claude helped solve two open physics problems and formalize proofs with 1,000 agents.Spezia news · 2026-11-10
  4. [4]Google DeepMind and Quantum AI researchers used AI agents and Lean proofs to cut quantum program resource costs.Spezia news · 2026-11-09
  5. [5]New ChemInsight benchmark finds top LLMs score poorly on chemistry reasoning, with Gemini 3 Pro leading.Spezia news · 2026-11-08
  6. [6]An unreleased Anthropic model proved the 20-year-old Aaronson–Ambainis quantum conjecture, verified by Lean.Spezia news · 2026-11-08
  7. [7]Tokyo Tech AI agents planned and ran quantum spin experiments, deriving a calibration law and confirming it in repeat measurements.Spezia news · 2026-11-06
  8. [8]China forum highlights AI-driven autonomous research agents, while experts stress human oversight in scientific discoverySpezia news · 2026-11-05
  9. [9]Physics - Add to calendar 11/10/2026 08:00 11/10/2026 17:00 Princeton University PCTS Workshop use-title Topic: AI-Accelerated Quantum Mat...Web
  10. [10]digest: 2026-11-03 daily digest · duanyytop/agents-radar@68372a0 - Original file line number | Diff line number | Diff line changeWeb · 2026-02-22

Other fields