UnifyBench

RankingEffort benchmarks

GPT-5.4 · Non-reasoning

Compare models

78 measured results · 6 effort settings · 54 benchmark/harness combinations

Model evidence and sources →

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
GPT-5.4 · Non-reasoning 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 context50100GPT-5.4 · Non-reasoning · Agentic: 4.2 · PreliminaryGPT-5.4 · Non-reasoning · Hard reasoning: 10.0 · SupportedGPT-5.4 · Non-reasoning · Coding: 27.9 · PreliminaryGPT-5.4 · Non-reasoning · Knowledge: 15.6 · SupportedGPT-5.4 · Non-reasoning · Multimodal: 14.3 · PreliminaryGPT-5.4 · Non-reasoning · Long context: 12.6 · Preliminary

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

  • Agentic

    A4.2Preliminary

  • Hard reasoning

    A10.0Supported

  • Coding

    A27.9Preliminary

  • Human pref

    Unknown

  • Knowledge

    A15.6Supported

  • Multimodal

    A14.3Preliminary

  • Long context

    A12.6Preliminary

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

GPT-5.4 · Non-reasoning: 5.1 overall comparison score. Meets the overall evidence requirements.

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

Which sources carry the weight?
  • Artificial Analysis · 100.0% direct comparison weight · independent of the model provider
Which benchmark families carry the weight?
  • GPQA · 11.8% direct comparison weight
  • IFBench · 11.8% direct comparison weight
  • Critpt · 11.8% direct comparison weight
  • Humanity’s Last Exam · 11.7% direct comparison weight
  • Mmmu Pro · 11.5% direct comparison weight
  • Other families (4) · 41.4% 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 2026-10-23.

Both ranks use the same surviving configurations for each removal. The cohort starts with 444 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 AnalysisComparison support lost; no rank203 remaining / 241 lostUnavailable
ARC Prize240 → 241Down 1444 remaining / 0 lost6.6 (+1.5 points)
Datacurve240 → 241Down 1444 remaining / 0 lost4.0 (-1.1 points)
LiveCodeBench240 → 243Down 3444 remaining / 0 lost5.0 (-0.1 points)
Scale240 → 242Down 2444 remaining / 0 lost5.9 (+0.8 points)
Terminal-Bench240 → 241Down 1444 remaining / 0 lost5.0 (-0.1 points)
Cognition240 → 245Down 5444 remaining / 0 lost4.9 (-0.2 points)
LiveBench239 → 239No movement443 remaining / 1 lost5.3 (+0.2 points)
SWE-rebench240 → 241Down 1444 remaining / 0 lost5.2 (+0.1 points)
Epoch AI234 → 236Down 2431 remaining / 13 lost5.1 (+0.0 points)
Vals AI240 → 240No movement443 remaining / 1 lost4.8 (-0.2 points)
Provider reports · Anthropic240 → 240No movement444 remaining / 0 lost5.1 (-0.0 points)
Provider reports · Moonshot AI240 → 241Down 1444 remaining / 0 lost4.9 (-0.2 points)
Provider reports · Meta240 → 241Down 1444 remaining / 0 lost5.0 (-0.0 points)
Provider reports · Mistral240 → 240No movement444 remaining / 0 lost5.1 (+0.0 points)
Provider reports · OpenAI240 → 240No movement444 remaining / 0 lost5.1 (+0.0 points)
All provider reports240 → 241Down 1444 remaining / 0 lost4.9 (-0.1 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-agi6.3+1.3
omniscience3.9-1.2
deepswe4.0-1.1
hle5.8+0.7
tau-bench5.8+0.7
terminal-bench4.5-0.6
epoch-game-puzzles4.7-0.3
gpqa5.3+0.3
simpleqa5.3+0.3
ifbench5.3+0.2
frontiercode4.9-0.2
mmlu-pro5.2+0.2
mmmu-pro5.2+0.1
gdp-pdf5.2+0.1
swe-rebench5.2+0.1
livebench-instructions5.2+0.1
multi-swe-bench5.0-0.1
critpt5.0-0.1
aa-lcr5.1+0.1
livebench-language5.1+0.1
arc-agi-35.0-0.1
terminal-bench-science5.1+0.1
apex-agents5.1+0.1
enigma-eval5.0-0.1
livebench-coding5.0-0.1
swe-atlas-qna5.0-0.1
harvey5.0-0.1
analyst-agent5.1+0.0
itbench5.0-0.0
livebench-reasoning5.0-0.0
vibe-code5.0-0.0
briefcase5.0-0.0
automationbench5.0-0.0
math5005.1+0.0
gmmlu5.1+0.0
swe-atlas-test-writing5.0-0.0
vals-code-migration5.0-0.0
vals-finance-agent5.0-0.0
vals-legal-research5.0-0.0
vals-tax-agent5.0-0.0
frontiermath5.1+0.0
aime5.1+0.0
gdpval5.0-0.0
vals-excel-modeling5.0-0.0
livebench-math5.1+0.0
livecodebench5.1+0.0
swe-atlas-refactoring5.1-0.0
enterprise-ops5.1-0.0
swe-bench-pro5.1+0.0
livebench-data5.1-0.0
What evidence is missing from the fit?

Across all collected settings for this model: 118 observations, 73 contributing, 0 matched without graph weight and 45 excluded. Counts do not establish rank eligibility.

  • 3 · Benchmark family or source protocol has not been reviewed
  • 1 · Effort is not identified
  • 3 · Evaluator has not been reviewed for independent admission
  • 20 · Cited comparator requires original evaluation provenance; not a new matched run.
  • 10 · Supporting evidence outside the reviewed capability core
  • 7 · No reviewed effort for this catalog observation; any separately collected effort measurement is counted separately
  • 1 · Microsoft report Table 11 and section 4.1: MAI Terminal-Bench removes timeouts and uses a minimal ReAct harness; comparator values are cited from other model releases. A common evaluation protocol is not established.
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.

Benchmark / sourceLowMediumHighXHighNon-reasoningUnspecified
APEX-Agents-AAArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
33.26%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

Not reportedNot reported
AA-LCR v1.1Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

76.67%
Reported settings

GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low

Not reportedNot reported
82%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

58.33%
Reported settings

GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning

Not reported
CritPtArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Benchmark developed by Argonne and UIUC, with contributions from 60+ researchers globally. CritPt is under review by its authors following external feedback · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

7.43%
Reported settings

GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low

Not reportedNot reported
23.43%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

0.57%
Reported settings

GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning

Not reported
GDPval-AA v2.1 LeaderboardArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Elo rating for performance on real-world work tasks · Anchored to DeepSeek V4.1 Flash (max) at 1600 · Higher is better · Evaluation results measured independently by Artificial Analysis

Unit: elo. Source collected 2026-10-18.

Not reportedNot reportedNot reported
1,246.23 Elo
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

Not reportedNot reported
GPQA DiamondArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

87.07%
Reported settings

GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low

Not reportedNot reported
92.02%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

74.85%
Reported settings

GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning

Not reported
Humanity's Last ExamArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

30.82%
Reported settings

GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low

Not reportedNot reported
43.74%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

11.26%
Reported settings

GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning

Not reported
IFBenchArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

65.92%
Reported settings

GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low

Not reportedNot reported
73.95%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

48.44%
Reported settings

GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning

Not reported
MMMU-ProArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

77.98%
Reported settings

GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low

Not reportedNot reported
78.44%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

70.58%
Reported settings

GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning

Not reported
AA-Omniscience IndexArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

AA-Omniscience Index (higher is better) measures knowledge reliability and hallucination. It rewards correct answers, penalizes hallucinations, and has no penalty for refusing to answer. Scores range from -100 to 100, where 0 means as many correct as incorrect answers, and negative scores mean more incorrect than correct. · Evaluation results measured independently by Artificial Analysis

Unit: index. Source collected 2026-10-18.

4.78
Reported settings

GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low

Not reportedNot reported
5.78
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

-15.67
Reported settings

GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning

Not reported
AA-Omniscience AccuracyArtificial Analysis · 6000 questions; 1 attempt; correct-answer share; GPT-5.6 Luna medium grader
Benchmark details

AA-Omniscience Accuracy (higher is better) measures the proportion of correctly answered questions out of all questions, regardless of whether the model chooses to answer · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

47.85%
Reported settings

GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low

Not reportedNot reported
50.85%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

37.38%
Reported settings

GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning

Not reported
𝜏²-Bench TelecomArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

74.56%
Reported settings

GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low

Not reportedNot reported
87.13%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

35.96%
Reported settings

GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning

Not reported
𝜏³-BankingArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Benchmark developed by Sierra Research · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
39.59%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

Not reportedNot reported
Terminal-Bench HardArtificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

43.18%
Reported settings

GPT-5.4 (Low); Model source: https://artificialanalysis.ai/models/gpt-5-4-low

Not reportedNot reported
57.58%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

37.88%
Reported settings

GPT-5.4 (Non-reasoning); Model source: https://artificialanalysis.ai/models/gpt-5-4-non-reasoning

Not reported
Terminal-Bench v2.1Artificial Analysis · Artificial Analysis published evaluation; see source for benchmark-specific settings
Benchmark details

Benchmark developed by the Laude Institute and open-source contributors · Independently benchmarked by Artificial Analysis · Evaluation results measured independently by Artificial Analysis

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
78.28%
Reported settings

GPT-5.4 (Xhigh); Model source: https://artificialanalysis.ai/models/gpt-5-4

Not reportedNot reported
ARC-AGI-2ARC Prize · v2 Semi-Private
Benchmark details

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

Unit: percent. Source collected 2026-10-18.

29.17%
Reported settings

GPT-5.4 (Low); Source model ID: gpt-5-4-low; 120 tasks

55.42%
Reported settings

GPT-5.4 (Medium); Source model ID: gpt-5-4-medium; 120 tasks

67.5%
Reported settings

GPT-5.4 (High); Source model ID: gpt-5-4-high; 120 tasks

73.95%
Reported settings

GPT-5.4 (XHigh); Source model ID: gpt-5-4-xhigh; 120 tasks

Not reportedNot reported
DeepSWE v1.1Datacurve · mini-swe-agent
Benchmark details

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

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
51.8%
Reported settings

gpt-5-4; Reported effort: xhigh; 113 unique tasks; 452 attempts; 4 runs

Not reportedNot reported
SWE-bench Pro publicScale · mini-swe-agent starred rows
Benchmark details

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

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
59.1%
Reported settings

gpt-5.4 (xHigh)*; 731 tasks; source date 2026-04-08

Not reportedNot reported
SWE Atlas · QnA · codexScale · codex
Benchmark details

124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
40.8%
Reported settings

124 code comprehension tasks; strict all-rubrics pass metric; Opus 4.5 judge; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-03-09T16:27:54.000Z; reported margin ±5.1 percentage points; No contamination annotation reported (not proof of absence)

Not reportedNot reported
SWE Atlas · Refactoring · codexScale · codex
Benchmark details

70 refactoring tasks across 10 repositories; behavior preservation and task-specific grading; source-reported agent harness.

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
44.29%
Reported settings

70 refactoring tasks across 10 repositories; behavior preservation and task-specific grading; source-reported agent harness.; published row 2026-05-06T01:44:16.000Z; reported margin ±6.76 percentage points; No contamination annotation reported (not proof of absence)

Not reportedNot reported
SWE Atlas · Test Writing · codexScale · codex
Benchmark details

90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
44.36%
Reported settings

90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-03-26T18:24:53.000Z; reported margin ±6.04 percentage points; No contamination annotation reported (not proof of absence)

Not reportedNot reported
SWE Atlas · Test Writing · mini-sweScale · mini-swe
Benchmark details

90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
40%
Reported settings

90 test-writing tasks; execution, mutation checks and rubric judging; source-reported agent harness. Repositories drawn from SWE-bench Pro; distinct tasks.; published row 2026-03-26T18:24:57.000Z; reported margin ±6 percentage points; No contamination annotation reported (not proof of absence)

Not reportedNot reported
LiveBench · AMPS Hard · 2026-06-25LiveBench · LiveBench official task evaluation
Benchmark details

LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
98%
Reported settings

LiveBench official task evaluation; livebench-2026-06-25-AMPS_Hard; gpt-5.4-xhigh; checkpoint=gpt-5.4

Not reportedNot reported
LiveBench · code completion · 2026-06-25LiveBench · LiveBench official task evaluation
Benchmark details

LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
80.44%
Reported settings

LiveBench official task evaluation; livebench-2026-06-25-code_completion; gpt-5.4-xhigh; checkpoint=gpt-5.4

Not reportedNot reported
LiveBench · code generation · 2026-06-25LiveBench · LiveBench official task evaluation
Benchmark details

LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
74.65%
Reported settings

LiveBench official task evaluation; livebench-2026-06-25-code_generation; gpt-5.4-xhigh; checkpoint=gpt-5.4

Not reportedNot reported
LiveBench · connections · 2026-06-25LiveBench · LiveBench official task evaluation
Benchmark details

LiveBench 2026-06-25 task snapshot. Task scores stay separate and related tasks share their reviewed family budget. Explicit checkpoint and effort labels only.

Unit: percent. Source collected 2026-10-18.

Not reportedNot reportedNot reported
100%
Reported settings

LiveBench official task evaluation; livebench-2026-06-25-connections; gpt-5.4-xhigh; checkpoint=gpt-5.4

Not 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 · 8,916 results across 259 models · Collected 2026-10-18