Checked Nov 7, 2026Start a new check

What was checked

Anthropic's Claude 4 just beat every human expert in medicine, law, and physics simultaneously in blind testing — 'it's over for professionals'

Assessment summary

Inaccurate

Anthropic’s system card reports mixed results: Claude Opus 4 reached or exceeded human baselines on three of four specified evaluations but performed substantially below the human baseline on FigQA. It does not report a single blind test in which Claude 4 beat every human expert across medicine, law, and physics.

Key findings

  • Anthropic's Claude 4 just beat every human expert in medicine, law, and physics simultaneously in blind testing

    Contradicted
    Anthropic’s system card reports mixed results: Claude Opus 4 reached or exceeded human baselines on three of four specified evaluations but performed substantially below the human baseline on FigQA. It does not report a single blind test in which Claude 4 beat every human expert across medicine, law, and physics.
    ChatGPT:
    Claude-family models have outperformed humans on several specific, narrow benchmarks and tasks (medicine and specialist exams), but there is no high-quality, publicly available blinded study showing Claude 4 beat every human expert across medicine, law, and physics at the same time. The claim conflates selective benchmark wins with a universal victory.
  • Claude 4 beat every human expert in medicine in blind testing

    Contradicted
    Claude Sonnet 4.6 scored 0.83 on Protocol QA, marginally above the human expert baseline of 0.79. This represents near-parity, not beating every expert.

    Only one AI assessment covered this finding.

  • 'It's over for professionals' - the implied claim that professional expertise is now obsolete

    Contradicted
    Sources consistently describe Claude as amplifying human expertise and requiring human validation, guidance, and oversight. No source supports the claim that professional expertise is now obsolete.

    Only one AI assessment covered this finding.

  • Claude 4 beat every human expert in law in blind testing

    Needs context
    Claude Opus 4.6 achieved 65.54% on law tasks in XpertBench testing, but this is not described as beating all human experts. The benchmark shows competitive performance, not dominance.

    Only one AI assessment covered this finding.

Model assessments

Assessment summary

  • The claim is contradicted by Anthropic’s own evaluations.
  • Claude Opus 4 reached or exceeded human baselines on some biology-related tests but performed substantially below the human baseline on another, and the available evidence does not show simultaneous blind testing against every human expert in medicine, law, and physics.www-cdn.anthropic.com The statement that professionals are finished is an unsupported exaggeration; Anthropic says humans can still outperform models given enough time.anthropic.com

Claims reviewed

Anthropic's Claude 4 just beat every human expert in medicine, law, and physics simultaneously in blind testing

Contradicted
Anthropic’s system card reports mixed results: Claude Opus 4 reached or exceeded human baselines on three of four specified evaluations but performed substantially below the human baseline on FigQA.www-cdn.anthropic.com It does not report a single blind test in which Claude 4 beat every human expert across medicine, law, and physics.

This claim may mislead readers without added context.

it's over for professionals

Needs context
This conclusion is not supported by the evidence. Anthropic states that human experts can still outperform current models when given sufficiently long time horizons.anthropic.com

This claim may mislead readers without added context.

Assessment summary

  • The claim contains multiple significant factual errors.
  • There is no model called "Claude 4"—current models are Opus 4.8, Sonnet 4.6, etc.
  • The evidence does not show Claude beating every human expert simultaneously in these domains; rather, it shows mixed performance with human involvement required.
  • The dramatic framing is unsupported by the actual research.

Claims reviewed

Claude 4 beat every human expert in medicine in blind testing

Contradicted
Claude Sonnet 4.6 scored 0.83 on Protocol QA, marginally above the human expert baseline of 0.79. This represents near-parity, not beating every expert.

This claim may mislead readers without added context.

'It's over for professionals' - the implied claim that professional expertise is now obsolete

Contradicted
Sources consistently describe Claude as amplifying human expertise and requiring human validation, guidance, and oversight. No source supports the claim that professional expertise is now obsolete.

This claim may mislead readers without added context.

Anthropic's Claude 4 just beat every human expert in medicine, law, and physics simultaneously

Contradicted
No model called 'Claude 4' exists currently; the latest models are Opus 4.8 and Sonnet 4.6. Evidence shows Claude performed marginally above human baseline in medicine (0.83 vs 0.79), achieved 65.54% on law tasks (not beating all experts), and solved physics problems with significant human guidance—not autonomous expert-beating.

This claim may mislead readers without added context.

Assessment summary

  • The headline overstates and generalizes selective benchmark results.
  • Claude (and other LLMs) have outperformed humans on some closed, narrowly scoped benchmarks in medicine, law, or physics, but no credible public evaluation shows Claude 4 beat every human expert across all three domains simultaneously in blinded, representative tests.
  • The claim that "it's over for professionals" is an unsupported value judgement and misleading.

Claims reviewed

Anthropic's Claude 4 just beat every human expert in medicine, law, and physics simultaneously in blind testing

Needs context
Claude-family models have outperformed humans on several specific, narrow benchmarks and tasks (medicine and specialist exams) anthropic.com ncbi.nlm.nih.gov beckershospitalreview.com, but there is no high-quality, publicly available blinded study showing Claude 4 beat every human expert across medicine, law, and physics at the same time. The claim conflates selective benchmark wins with a universal victory.

This claim may mislead readers without added context.

'it's over for professionals' (i.e., professionals are now obsolete because Claude 4 beat human experts)

Can’t verify
This is an opinion and prediction rather than a verifiable factual statement. Existing evaluations show LLMs can exceed humans on some tasks but do not demonstrate broad, reliable, real-world replacement of professionals anthropic.com anthropic.com source unavailable.

Sources cited

18 sources

Share this result