RankingEffort benchmarks

Claude Opus 4

Compare models

9 measured results · 4 effort settings · 6 benchmark/harness combinations

Change model

Compare with

Choose a second model and its effort to see both profiles and benchmark differences below.

Compare with

Capability profile

One configuration, seven capabilities

Comparison score · 0–100
Claude Opus 4 · Thinking capability radarComparison scores from 0 to 100. Unknown capabilities have no point; lines stop at gaps. Hollow points are preliminary. Exact values and support labels follow the chart.AgenticHard reasoningCodingHuman prefKnowledgeMultimodalLong context50100Claude Opus 4 · Thinking · Hard reasoning: 7.2 · SupportedClaude Opus 4 · Thinking · Coding: 0.6 · Preliminary

Gaps are unknown, not zero. Hollow points are preliminary.

  • Agentic

    Unknown
  • Hard reasoning

    A7.2Supported
  • Coding

    A0.6Preliminary
  • Human pref

    Unknown
  • Knowledge

    Unknown
  • Multimodal

    Unknown
  • Long context

    Unknown

Each profile belongs to the named configuration. Capability scores use a shared panel for all axes, distinct from the overall ranking panel; they are not accuracy percentages or direct head-to-head wins.

Capability score evidence

Claude Opus 4 · Thinking: 1.2 overall comparison score. Meets the overall evidence requirements.

2 areas measured. The available task mix affects this estimate; missing areas remain unknown.

Which sources carry the weight?
  • LiveCodeBench · 39.5% direct comparison weight · independent of the model provider
  • Scale · 30.9% direct comparison weight · independent of the model provider
  • Epoch AI · 29.6% direct comparison weight · independent of the model provider
Which benchmark families carry the weight?
  • LiveCodeBench · 39.5% direct comparison weight
  • Humanity’s Last Exam · 30.9% direct comparison weight
  • GPQA · 29.6% direct comparison weight

These shares describe direct comparison weight, not fractions of the final score or confidence. Source balancing does not give every family equal weight or remove correlations between overlapping sources.

How sensitive is this score?

The audit removes one source at a time and refits against the same overall reference panel. These published results use the matching dataset audited 2026-11-22.

Both ranks use the same surviving configurations for each removal. The cohort starts with 602 configurations supported in the original audit, including separate effort settings. These are diagnostic positions, not the default leaderboard ranks. Eligibility is not rechecked after removal.

What changes when a source is removed?
Removed sourceRank before → afterRank movementCohort remaining / lost supportScore after (change)
Artificial Analysis219 → 215Up 4230 remaining / 372 lost1.2 (-0.1 points)
ARC Prize435 → 555Down 120602 remaining / 0 lost0.3 (-1.0 points)
Datacurve435 → 435No movement602 remaining / 0 lost1.0 (-0.3 points)
LiveCodeBench435 → 407Up 28602 remaining / 0 lost1.8 (+0.5 points)
Scale435 → 443Down 8602 remaining / 0 lost1.3 (+0.1 points)
Terminal-Bench435 → 435No movement602 remaining / 0 lost1.2 (-0.0 points)
Cognition435 → 433Up 2602 remaining / 0 lost1.2 (-0.0 points)
LiveBench434 → 434No movement601 remaining / 1 lost1.3 (+0.1 points)
SWE-rebench435 → 437Down 2602 remaining / 0 lost1.2 (-0.0 points)
Epoch AI421 → 482Down 61588 remaining / 14 lost0.5 (-0.7 points)
Vals AI434 → 436Down 2601 remaining / 1 lost1.2 (-0.1 points)
Provider reports · Anthropic435 → 435No movement602 remaining / 0 lost1.2 (-0.0 points)
Provider reports · Moonshot AI435 → 434Up 1602 remaining / 0 lost1.2 (-0.0 points)
Provider reports · Meta435 → 434Up 1602 remaining / 0 lost1.2 (-0.0 points)
Provider reports · Mistral435 → 435No movement602 remaining / 0 lost1.2 (+0.0 points)
Provider reports · OpenAI435 → 435No movement602 remaining / 0 lost1.3 (+0.0 points)
All provider reports435 → 434Up 1602 remaining / 0 lost1.2 (-0.0 points)

Lost support means a configuration no longer has a comparison path to the complete reference panel. Its rank is unavailable. A surviving path alone does not establish leaderboard eligibility.

Overlapping sources may remain correlated even with source balancing. Leave-one-source-out movement is a sensitivity diagnostic, not a spread of source-specific ranks or a confidence measure. Scoring methodology · Dated configuration audit

Every benchmark family
Removed evidenceRecomputed scoreChange
arc-agi0.3-1.0
gpqa0.5-0.7
livecodebench1.8+0.5
simpleqa1.6+0.3
deepswe1.0-0.3
aime1.4+0.2
epoch-game-puzzles1.1-0.1
mmmu-pro1.3+0.1
arc-agi-31.2-0.1
mmlu-pro1.3+0.0
frontiercode1.2-0.0
livebench-instructions1.3+0.0
hle1.2-0.0
multi-swe-bench1.2-0.0
swe-bench-pro1.3+0.0
frontiermath1.3+0.0
livebench-language1.3+0.0
aa-lcr1.2-0.0
livebench-coding1.2-0.0
swe-atlas-refactoring1.3+0.0
gmmlu1.2-0.0
enigma-eval1.3+0.0
vibe-code1.2-0.0
swe-atlas-qna1.2-0.0
terminal-bench1.2-0.0
apex-agents1.3+0.0
vals-code-migration1.2-0.0
automationbench1.2-0.0
livebench-reasoning1.2-0.0
briefcase1.2-0.0
vals-finance-agent1.2-0.0
gdp-pdf1.3+0.0
vals-legal-research1.2-0.0
vals-excel-modeling1.2-0.0
tau-bench1.3+0.0
vals-tax-agent1.2-0.0
itbench1.2-0.0
swe-atlas-test-writing1.2-0.0
gdpval1.3+0.0
analyst-agent1.3+0.0
ifbench1.2-0.0
omniscience1.2-0.0
terminal-bench-science1.3+0.0
enterprise-ops1.2-0.0
swe-rebench1.2-0.0
math5001.2-0.0
livebench-data1.2+0.0
livebench-math1.2+0.0
harvey1.2+0.0
critpt1.2+0.0
What evidence is missing from the fit?

Across all collected settings for this model: 9 observations, 7 contributing, 0 matched without graph weight and 2 excluded. Counts do not establish rank eligibility.

  • 2 · Conflicting values for this configuration in the same evaluation
Inspect every observation and exclusion →

Benchmark results

These are published benchmark scores. Supported effort estimates also appear on the Capability leaderboard. Results from different harnesses occupy different rows. Missing cells stay unknown; no scores are copied between effort levels.

Reviewed retained observations are in use. October 2026 Real-SWE board has no 95% confidence whisker, so these scores stay at the 2026-10-18 retrieval. Original source retrieval 2026-10-18; source updated 2026-10-18.

Benchmark / sourceThinkingThinking 1KThinking 8KThinking 16K
ARC-AGI-2ARC Prize · v2 Semi-Private
Benchmark details

Original public board. Configuration and task coverage may differ; inspect source.

Unit: percent. Source collected 2026-11-22.

Not reported
0%
Reported settings

Claude Opus 4 (Thinking 1K); Source model ID: claude-opus-4-20250514-thinking-1k; 120 tasks

4.52%
Reported settings

Claude Opus 4 (Thinking 8K); Source model ID: claude-opus-4-20250514-thinking-8k; 120 tasks

8.61%
Reported settings

Claude Opus 4 (Thinking 16K); Source model ID: claude-opus-4-20250514-thinking-16k; 120 tasks

LiveCodeBenchLiveCodeBench · generation pass@1
Benchmark details

Evaluation window 2024-08-01 to 2025-05-01; 454 shared tasks.

Unit: percent. Source collected 2026-11-22.

56.6%
Reported settings

Claude-Opus-4 (Thinking); Exact source model: claude-opus-4-20250514; 454 tasks

Not reportedNot reportedNot reported
HLE · text only · Scale official evaluationScale · Scale official evaluation
Benchmark details

Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.

Unit: percent. Source collected 2026-11-22.

10.8%
Reported settings

Scale text-only public HLE subset; shares the full HLE family budget; automatic answer judging.; published row 2025-05-24T06:34:16.000Z; reported margin ±1.31 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions. Sampled at temperature 0.3.

Not reportedNot reportedNot reported
HLE · full · Scale official evaluationScale · Scale official evaluation
Benchmark details

Scale full public HLE; 2,500 questions; automatic answer judging; temperature 0 when configurable. Public-set contamination caveats retained per result.

Unit: percent. Source collected 2026-11-22.

10.72%
Reported settings

Scale full public HLE; 2,500 questions; automatic answer judging; temperature 0 when configurable. Public-set contamination caveats retained per result.; published row 2025-05-24T06:35:10.000Z; reported margin ±1.21 percentage points; Potential contamination warning: This model was evaluated after the public release of HLE, allowing model builder access to the prompts and solutions.

Not reportedNot reportedNot reported
Epoch · GPQA diamond · v1.0.0Epoch AI · Epoch AI Inspect; task 1.0.0; scorer choice
Benchmark details

Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.

Unit: percent. Source collected 2026-11-22.

76.26%
Reported settings

Epoch AI Inspect; task 1.0.0; scorer choice; model=claude-opus-4-20250514_16K; run=CtxFusPxCraL3EeRJu2c5k; mean accuracy; https://epoch.ai/data/benchmarks.csv

Not reportedNot reportedNot reported
Epoch · OTIS Mock AIME 2024-2025 · v1.0.0Epoch AI · Epoch AI Inspect; task 1.0.0; scorer model_graded
Benchmark details

Epoch AI internal evaluation. Mean accuracy under the published setting; exact task version and scorer stay separate. Related tasks share a family budget. Data by Epoch AI, CC BY.

Unit: percent. Source collected 2026-11-22.

64.44%
Reported settings

Epoch AI Inspect; task 1.0.0; scorer model_graded; model=claude-opus-4-20250514_27K; run=VrRccga4uRVuCjNp637MeY; mean accuracy; https://epoch.ai/data/benchmarks.csv

60%
Reported settings

Epoch AI Inspect; task 1.0.0; scorer model_graded; model=claude-opus-4-20250514_16K; run=Xz2F28L5k8HFNfY7Lo8Ggw; mean accuracy; https://epoch.ai/data/benchmarks.csv

Not reportedNot reportedNot reported
What these comparisons mean

Each source reports its own effort labels. Matching labels do not establish equal compute budgets or identical fallback behavior. Open Reported settings to inspect the exact model name, agent and task coverage. Multiple results at one level remain visible rather than selecting the best. Source-reported None or Non-reasoning labels are not treated as evidence that an API supports those settings.

These measurements use their own dated source collection. The leaderboard fits identified configurations jointly; it contains no pooled model ratings. Explicit thinking budgets are separate configurations. Read the methodology.

Download all effort benchmark results · 10,184 results across 381 models · Collected 2026-11-22