Benchmarks/Anthropic Key Life Sciences Evals

suite · audited · verified 2026-07-22

Anthropic Key Life Sciences Evals

An Anthropic private internal suite reported only through an official accuracy chart covering scientific figure interpretation, computational biology, and protein understanding.

Private/internal benchmark: tasks, counts, and core protocol are not public. This record is excluded from public and runnable coverage statistics. Exact labeled deltas remain inspectable, but are not converted into absolute scores or plotted as a ranking.

Benchmark definition

What is counted

Version: reported-2026-01-11
Total: Not reported (internal evaluation tasks shown only as three directions in an official chart)
Task formats: private internal accuracy evaluation; exact format not reported
Capabilities: KnowledgeEvidence synthesisData analysisScientific reasoning
Modalities: TextFigure

Version history

Version	Status	Release / as-of	Total	Formal tracks
reported-2026-01-11 `anthropic-key-life-sciences-reported-2026-01-11`	current	2026-01-11	Not reported (internal evaluation tasks shown only as three directions in an official chart)	anthropic-scientific-figure-interpretation, anthropic-computational-biology, anthropic-protein-understanding

Registered child tracks

Anthropic Computational Biology Eval

Private Anthropic computational-biology evaluation direction reported only through a model-trend chart, without public tasks, counts, or protocol details.

1 evaluation run(s)

Anthropic Protein Understanding Eval

Private Anthropic protein-understanding evaluation direction reported only through a model-trend chart, without public tasks, counts, or protocol details.

1 evaluation run(s)

Anthropic Scientific Figure Interpretation Eval

Private Anthropic evaluation direction for scientific figure interpretation, reported only through a model-trend chart with no task count or released examples.

1 evaluation run(s)

Scientific Task Atlas

Scientific task classification

partial for reported-2026-01-11 · as of 2026-01-11. The official chart names three private directions but does not disclose tasks or an exhaustive scientific taxonomy.

The source names only a broad direction; no more specific leaf task can be assigned without inference.

Scientific coverage notes

Domain	Coverage	Count	Interpretation
Life science	explicitly-in-scope	Not reported	Three named internal directions are public, but tasks and counts are not.

Relationship registry

How works use this benchmark

Partial claims, non-evaluation uses, and third-party summaries stay visible without entering model comparisons.

Normalized evaluations

evaluation

anthropic-computational-biology-evaluation

Benchmark: Anthropic Computational Biology Eval · version reported-2026-01-11

Work: Advancing Claude in healthcare and the life sciences · source version anthropic-healthcare-life-sciences-2026-01-11

Selection: not reported
Models: Claude Opus 4.1, Claude Sonnet 4.5, Claude Opus 4.5
Metrics: Accuracy delta
Linked runs: anthropic-computational-biology-delta

Not reported / unresolved: task count; task data; grader; prompt; tools; repeats; absolute accuracy

Only the exact Opus 4.5 improvement annotation is normalized; plotted absolute values are not estimated.

Evidence

figure: Evals for key life sciences tasks — Computational biology (Shows Opus 4.1, Sonnet 4.5, Opus 4.5 and an exact +10.5% annotation.)
Supports: /benchmark_version, /relation_type, /status, /model_ids, /scope, /metric_labels, /evaluation_run_ids, /reporting_gaps, /notes

evaluation

anthropic-protein-understanding-evaluation

Benchmark: Anthropic Protein Understanding Eval · version reported-2026-01-11

Work: Advancing Claude in healthcare and the life sciences · source version anthropic-healthcare-life-sciences-2026-01-11

Selection: not reported
Models: Claude Opus 4.1, Claude Sonnet 4.5, Claude Opus 4.5
Metrics: Accuracy delta
Linked runs: anthropic-protein-understanding-delta

Not reported / unresolved: task count; task data; grader; prompt; tools; repeats; absolute accuracy

Only the exact Opus 4.5 improvement annotation is normalized; plotted absolute values are not estimated.

Evidence

figure: Evals for key life sciences tasks — Protein understanding (Shows Opus 4.1, Sonnet 4.5, Opus 4.5 and an exact +10.3% annotation.)
Supports: /benchmark_version, /relation_type, /status, /model_ids, /scope, /metric_labels, /evaluation_run_ids, /reporting_gaps, /notes

evaluation

anthropic-scientific-figure-evaluation

Benchmark: Anthropic Scientific Figure Interpretation Eval · version reported-2026-01-11

Work: Advancing Claude in healthcare and the life sciences · source version anthropic-healthcare-life-sciences-2026-01-11

Selection: not reported
Models: Claude Opus 4.1, Claude Sonnet 4.5, Claude Opus 4.5
Metrics: Accuracy delta
Linked runs: anthropic-scientific-figure-delta

Not reported / unresolved: task count; task data; grader; prompt; tools; repeats; absolute accuracy

Only the exact Opus 4.5 improvement annotation is normalized; plotted absolute values are not estimated.

Evidence

figure: Evals for key life sciences tasks — Scientific figure interpretation (Shows Opus 4.1, Sonnet 4.5, Opus 4.5 and an exact +13.2% annotation.)
Supports: /benchmark_version, /relation_type, /status, /model_ids, /scope, /metric_labels, /evaluation_run_ids, /reporting_gaps, /notes

Creation, training, validation, or model-selection uses

benchmark creation

anthropic-key-life-sciences-creation

Work: Advancing Claude in healthcare and the life sciences · source version anthropic-healthcare-life-sciences-2026-01-11

Selection: not applicable
Models: Not reported / not applicable
Metrics: Not reported / not applicable
Linked runs: None

Anthropic is the creator and evaluator of this private internal suite; child-track evaluation relations are recorded separately.

Evidence

figure: Evals for key life sciences tasks (Anthropic page presents the three private internal directions and model trend lines.)
Supports: /benchmark_version, /relation_type, /status, /scope, /notes

Evaluation registry

Works and run settings

A setting change—scope, prompt, tools, budget, grader, or repeats—creates a separate run. Charts never cross a comparability group.

Evaluation run

anthropic-computational-biology-delta

From Advancing Claude in healthcare and the life sciences

Evaluated models / systems: Claude Opus 4.1, Claude Opus 4.5, Claude Sonnet 4.5

Scopeunknown

ShotsNot reported

TurnsNot reported

System prompt publicNot reported

Reasoning / effortNot reported

BrowserNot reported

InternetNot reported

DatabasesNot reported

Code executionNot reported

ContainerNot reported

External toolsNot reported

Token budgetNot reported

Time / cost budgetNot reported

TemperatureNot reported

SeedNot reported

RepeatsNot reported

GraderNot reported

StatisticsNot reported

ContaminationNot reported

Metrics, results, and full protocol

Metrics

Metric	Kind / baseline	Unit	Aggregation	Threshold / tolerance
Accuracy improvement from Claude Opus 4.1 to Claude Opus 4.5	delta baseline: Claude Opus 4.1	percent delta as annotated	Not reported	Not reported

Results

Model	Metric	Value	n
Claude Opus 4.5	Accuracy improvement from Claude Opus 4.1 to Claude Opus 4.5	Δ 10.5 percent delta as annotated Exact +10.5% chart annotation; no absolute accuracy or relative-versus-percentage-point interpretation is inferred.	Not reported

Evidence

figure: Evals for key life sciences tasks — Computational biology (Labels Accuracy (%) and annotates +10.5% from Opus 4.1 to Opus 4.5.) — supports /benchmark_version, /model_ids, /scope, /protocol, /metrics, /results

Evaluation run

anthropic-protein-understanding-delta

From Advancing Claude in healthcare and the life sciences

Evaluated models / systems: Claude Opus 4.1, Claude Opus 4.5, Claude Sonnet 4.5

Scopeunknown

ShotsNot reported

TurnsNot reported

System prompt publicNot reported

Reasoning / effortNot reported

BrowserNot reported

InternetNot reported

DatabasesNot reported

Code executionNot reported

ContainerNot reported

External toolsNot reported

Token budgetNot reported

Time / cost budgetNot reported

TemperatureNot reported

SeedNot reported

RepeatsNot reported

GraderNot reported

StatisticsNot reported

ContaminationNot reported

Metrics, results, and full protocol

Metrics

Metric	Kind / baseline	Unit	Aggregation	Threshold / tolerance
Accuracy improvement from Claude Opus 4.1 to Claude Opus 4.5	delta baseline: Claude Opus 4.1	percent delta as annotated	Not reported	Not reported

Results

Model	Metric	Value	n
Claude Opus 4.5	Accuracy improvement from Claude Opus 4.1 to Claude Opus 4.5	Δ 10.3 percent delta as annotated Exact +10.3% chart annotation; no absolute accuracy or relative-versus-percentage-point interpretation is inferred.	Not reported

Evidence

figure: Evals for key life sciences tasks — Protein understanding (Labels Accuracy (%) and annotates +10.3% from Opus 4.1 to Opus 4.5.) — supports /benchmark_version, /model_ids, /scope, /protocol, /metrics, /results

Evaluation run

anthropic-scientific-figure-delta

From Advancing Claude in healthcare and the life sciences

Evaluated models / systems: Claude Opus 4.1, Claude Opus 4.5, Claude Sonnet 4.5

Scopeunknown

ShotsNot reported

TurnsNot reported

System prompt publicNot reported

Reasoning / effortNot reported

BrowserNot reported

InternetNot reported

DatabasesNot reported

Code executionNot reported

ContainerNot reported

External toolsNot reported

Token budgetNot reported

Time / cost budgetNot reported

TemperatureNot reported

SeedNot reported

RepeatsNot reported

GraderNot reported

StatisticsNot reported

ContaminationNot reported

Metrics, results, and full protocol

Metrics

Metric	Kind / baseline	Unit	Aggregation	Threshold / tolerance
Accuracy improvement from Claude Opus 4.1 to Claude Opus 4.5	delta baseline: Claude Opus 4.1	percent delta as annotated	Not reported	Not reported

Results

Model	Metric	Value	n
Claude Opus 4.5	Accuracy improvement from Claude Opus 4.1 to Claude Opus 4.5	Δ 13.2 percent delta as annotated Exact +13.2% chart annotation; no absolute accuracy or relative-versus-percentage-point interpretation is inferred.	Not reported

Evidence

figure: Evals for key life sciences tasks — Scientific figure interpretation (Labels Accuracy (%) and annotates +13.2% from Opus 4.1 to Opus 4.5.) — supports /benchmark_version, /model_ids, /scope, /protocol, /metrics, /results

Evidence and change history

Source locators remain visible; expand an item to inspect the exact Registry fields it supports.

Advancing Claude in healthcare and the life sciences · figure: Evals for key life sciences tasks (Names the three internal evaluation directions, labels Accuracy (%), and annotates Opus 4.5 changes from Opus 4.1.) · Supports 30 fields

Open source →

/name
/aliases
/summary
/kind
/parent_id
/organizations
/release_date
/latest_version
/domains
/capabilities
/modalities
/task_formats
/task_counts/total
/task_counts/basis
/task_counts/subsets
/coverage_notes
/access/level
/access/tasks
/access/artifacts
/access/grader
/access/license
/access/biosafety_notes
/resources
/implementations
/versions/0/release_date
/versions/0/as_of
/versions/0/task_counts/total
/versions/0/task_counts/basis
/versions/0/task_counts/subsets
/versions/0/formal_tracks

View source-level modification history on GitHub →