WhitzardNUWA Lab ↗
NUWANUWA INDEX

Search NUWA Index

Research · Benchmarks · Findings · Perspectives

中文
← Models

OpenAI

GPT-5.4

Published evidence on capabilities, behavioral tendencies and observed actions.

First public availability
2026-03-05
Parameters
Not disclosed
Metadata updated
2026-10-27
Sources ↓

Evidence overview

Published risk evidence is organized by assessment and protocol. Missing evidence is not a low-risk finding.

NUWA curates and interprets these sources. Experiments are attributed to their original executors; this page does not claim NUWA has independently replicated them.

MODEL MONITORING

Representative benchmark coverage

Results by protocol and configuration; gaps show records awaiting collection or linkage.

Cybench →

No eligible measurement linked to this model version.

CyberGym →

3 measurements

  • 66.3%Not disclosed · Codex CLI
  • 73.8%Not disclosed · VARAS-OneMind \(智安·宇盾\)
  • 79%Not disclosed · OpenAI Agent

METR Time Horizons →

2 measurements

  • 341.74 minMETR · mtb/start_metr_task,metr_agents/react; ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react; re_bench_common/init_solver,use_tools,metr_agents/react; metr_tasks_swaa/init_solver,generate; metr_agents/react
  • 53.88 minMETR · mtb/start_metr_task,metr_agents/react; ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react; re_bench_common/init_solver,use_tools,metr_agents/react; metr_tasks_swaa/init_solver,generate; metr_agents/react

MASK →

No eligible measurement linked to this model version.

All linked measurements

Each result retains its protocol and unit.

22 observations

Measurements and source conditions
System / modelResultCondition statusTask / conditionsEvaluator / reportDates
GPT-5.4 · Codex CLIGPT-5.466.3%CyberGym · Success rateIncompletelevel1score_10Codex CLINot disclosedCyberGym official leaderboard and machine-data extractionRelease 2026-03-05Report —
GPT-5.4 · VARAS-OneMind \(智安·宇盾\)GPT-5.473.8%CyberGym · Success rateIncompletelevel1score_10VARAS-OneMind \(智安·宇盾\)Not disclosedCyberGym official leaderboard and machine-data extractionRelease 2026-03-05Report —
GPT-5.4 · OpenAI AgentGPT-5.479%CyberGym · Success rateIncompletelevel1score_10OpenAI AgentNot disclosedCyberGym official leaderboard and machine-data extractionRelease 2026-03-05Report —
GPT-5.40%Agentic misalignment · deliberate harmful action rateIncompleteAnthropic Agentic Misalignmentdeliberate harmful action rate—MetaEvaluating DeepSeek v4 Pro for Frontier RisksRelease 2026-03-05Report 2026-05-29
GPT-5.41.2%DeceptionBench · deception score: single-turnIncompleteDeceptionBenchdeception score: single-turn—MetaEvaluating DeepSeek v4 Pro for Frontier RisksRelease 2026-03-05Report 2026-05-29
GPT-5.41.5%DeceptionBench · deception score: multi-turn w pressureIncompleteDeceptionBenchdeception score: multi-turn w pressure—MetaEvaluating DeepSeek v4 Pro for Frontier RisksRelease 2026-03-05Report 2026-05-29
GPT-5.41.2%DeceptionBench · DeceptionBench deception scoreIncompleteDeceptionBenchsingle-turn—MetaEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion RisksRelease 2026-03-05Report 2026-07-03
GPT-5.41.5%DeceptionBench · DeceptionBench deception scoreIncompleteDeceptionBenchmulti-turn with pressure—MetaEarly Results on Evaluating GLM 5.2 for Behavioral and Persuasion RisksRelease 2026-03-05Report 2026-07-03
GPT-5.454.4%CTF-Archive-Diamond · benchmark_scoreIncompleteCTF-Archive Diamondxhigh1M weighted tokensInspect built-in ReAct agentNIST / CAISICAISI Evaluation of DeepSeek V4 ProRelease 2026-03-05Report 2026-05-01
GPT-5.4 · METR scaffoldGPT-5.4341.74 minMETR Time Horizons · 50% task-completion horizonSource 95% CI [186.581591, 768.779526] minIncompleteMETR-Horizon-v1.1mtb/start_metr_task,metr_agents/react; ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react; re_bench_common/init_solver,use_tools,metr_agents/react; metr_tasks_swaa/init_solver,generate; metr_agents/reactMETRMETR Time Horizon 1.1 — source dataRelease 2026-03-05Report —
GPT-5.4 · METR scaffoldGPT-5.453.88 minMETR Time Horizons · 80% task-completion horizonSource 95% CI [23.957027, 108.679232] minIncompleteMETR-Horizon-v1.1mtb/start_metr_task,metr_agents/react; ai_rd_fix_embedding/init_solver,use_tools,metr_agents/react; re_bench_common/init_solver,use_tools,metr_agents/react; metr_tasks_swaa/init_solver,generate; metr_agents/reactMETRMETR Time Horizon 1.1 — source dataRelease 2026-03-05Report —
GPT-5.4 · Codex CLIGPT-5.461 countExploitGym · Intended-target successesIncompletev1on_target2h timeoutCodex CLINot disclosedExploitGym official leaderboard and benchmark pageRelease 2026-03-05Report —
GPT-5.439%SEC-bench Pro · Headline success rateIncomplete260617 / overallheadlineHeadline UI: fixed budget; Completed UI: within timeout. Published summary timeout_caps=null; cap labels do not establish equal token or cost budgets.CodexNot disclosedSEC-bench Pro official leaderboard: selected 260617 and historical 260505Release 2026-03-05Report —
GPT-5.440.5%SEC-bench Pro · Completed-mode success rateIncomplete260617 / overallcompletedHeadline UI: fixed budget; Completed UI: within timeout. Published summary timeout_caps=null; cap labels do not establish equal token or cost budgets.CodexNot disclosedSEC-bench Pro official leaderboard: selected 260617 and historical 260505Release 2026-03-05Report —
GPT-5.435%SEC-bench Pro · Headline success rateIncomplete260617 / v8headlineHeadline UI: fixed budget; Completed UI: within timeout. Published summary timeout_caps=["1.3h"]; cap labels do not establish equal token or cost budgets.CodexNot disclosedSEC-bench Pro official leaderboard: selected 260617 and historical 260505Release 2026-03-05Report —
GPT-5.436.7%SEC-bench Pro · Completed-mode success rateIncomplete260617 / v8completedHeadline UI: fixed budget; Completed UI: within timeout. Published summary timeout_caps=["1.3h"]; cap labels do not establish equal token or cost budgets.CodexNot disclosedSEC-bench Pro official leaderboard: selected 260617 and historical 260505Release 2026-03-05Report —
GPT-5.425%SEC-bench Pro · Headline success rateIncomplete260617 / firefoxheadlineHeadline UI: fixed budget; Completed UI: within timeout. Published summary timeout_caps=[]; cap labels do not establish equal token or cost budgets.CodexNot disclosedSEC-bench Pro official leaderboard: selected 260617 and historical 260505Release 2026-03-05Report —
GPT-5.425%SEC-bench Pro · Completed-mode success rateIncomplete260617 / firefoxcompletedHeadline UI: fixed budget; Completed UI: within timeout. Published summary timeout_caps=[]; cap labels do not establish equal token or cost budgets.CodexNot disclosedSEC-bench Pro official leaderboard: selected 260617 and historical 260505Release 2026-03-05Report —
GPT-5.452.6%SEC-bench Pro · Headline success rateIncomplete260617 / linuxheadlineHeadline UI: fixed budget; Completed UI: within timeout. Published summary timeout_caps=["1.5h"]; cap labels do not establish equal token or cost budgets.CodexNot disclosedSEC-bench Pro official leaderboard: selected 260617 and historical 260505Release 2026-03-05Report —
GPT-5.456.5%SEC-bench Pro · Completed-mode success rateIncomplete260617 / linuxcompletedHeadline UI: fixed budget; Completed UI: within timeout. Published summary timeout_caps=["1.5h"]; cap labels do not establish equal token or cost budgets.CodexNot disclosedSEC-bench Pro official leaderboard: selected 260617 and historical 260505Release 2026-03-05Report —
GPT-5.40.4 percentage pointsAlignment faking · Compliance gapIncompleteAlignment FakingCompliance gap—MetaEvaluating DeepSeek v4 Pro for Frontier RisksRelease 2026-03-05Report 2026-05-29
GPT-5.40%Alignment faking · Deceptive alignment rateIncompleteAlignment FakingDeceptive alignment rate—MetaEvaluating DeepSeek v4 Pro for Frontier RisksRelease 2026-03-05Report 2026-05-29

Developer disclosures

No developer-authored safety report has been linked to this exact model record in the current collection.

External assessments

SaferAI

GLM-5.2 Risk Evaluation Report →

The report finds substantial task capability and pressure-dependent behavior; comparison figures include both SaferAI runs and reused external results.

2026-08-02

METR

METR Time Horizon 1.1 →

Estimated task difficulty at 50% and 80% success, measured in human-expert minutes.

Publication date not established

CyberGym official snapshot →

Published evaluation results collected from CyberGym, retaining the reported systems and configurations.

Publication date not established

ExploitGym official snapshot →

Published evaluation results collected from ExploitGym, retaining the reported systems and configurations.

Publication date not established

SEC-bench Pro official snapshot →

Published evaluation results collected from SEC-bench Pro, retaining the reported systems and configurations.

Publication date not established

Evidence by risk topic

Autonomy 2 protocol results

METR Time Horizon 1.1 →

External evaluation · Executed by METR

50% task-completion horizon
341.74 min
80% task-completion horizon
53.88 min

Task counts/weights vary with source tasks and scaffold coverage; see methodology.

Agentic Misalignment →

Reused result · Executed by Meta · Original executor Meta

deliberate harmful action rate
0%

Per-result sample size not disclosed

Deception 7 protocol results

DeceptionBench →

Reused result · Executed by Meta · Original executor Meta

deception score: single-turn
1.2%

Per-result sample size not disclosed

DeceptionBench →

Reused result · Executed by Meta · Original executor Meta

deception score: multi-turn w pressure
1.5%

Per-result sample size not disclosed

Alignment faking →

Reused result · Executed by Meta · Original executor Meta

Compliance gap
0.4 percentage points

Per-result sample size not disclosed

Alignment faking →

Reused result · Executed by Meta · Original executor Meta

Deceptive alignment rate
0%

Per-result sample size not disclosed

Agentic Misalignment →

Reused result · Executed by Meta · Original executor Meta

deliberate harmful action rate
0%

Per-result sample size not disclosed

DeceptionBench →

Reused result · Executed by Meta · Original executor Meta

DeceptionBench deception score
1.2%

Per-result sample size not disclosed

DeceptionBench →

Reused result · Executed by Meta · Original executor Meta

DeceptionBench deception score
1.5%

Per-result sample size not disclosed

Cybersecurity 13 protocol results

CTF-Archive-Diamond →

External evaluation · Executed by NIST / CAISI

benchmark_score
54.4%

Per-result sample size not disclosed

CBRN 1 protocol results

WMDP →

Reused result · Executed by Meta · Original executor Meta

accuracy
80.1%

Per-result sample size not disclosed

Disinformation No structured result linked

A gap in this collection is not evidence of low risk. Consult the linked reports for unextracted findings and protocol scope.

Autonomous physical harm No structured result linked

A gap in this collection is not evidence of low risk. Consult the linked reports for unextracted findings and protocol scope.

Source review and evidence gaps

CyberGym →

Published evaluation results collected from CyberGym, retaining the reported systems and configurations.

The site reports submitted systems under different configurations. Per-row task counts, budgets and execution dates are not fully disclosed. Undefined metrics and unresolved identity rows are excluded. Raw JSON fractions are shown as percentages using ×100 and one-decimal rounding, matching the official renderer.

ExploitGym →

Published evaluation results collected from ExploitGym, retaining the reported systems and configurations.

Intended-target success and flag capture are distinct counts. Task versions and timeout cutoffs remain separate. Catalogue size is not assumed to be the denominator for each submission.

SEC-bench Pro →

Published evaluation results collected from SEC-bench Pro, retaining the reported systems and configurations.

Headline and Completed apply different inclusion rules and may change both numerator and denominator. Provider and backend labels do not identify the actual evaluator.

Evaluating DeepSeek v4 Pro for Frontier Risks →

Source-specific extraction of Neo Research · DeepSeek V4 Pro. Findings and measurements retain their original attribution.

Source transcription is not experimental replication. Unknown dates, evaluator roles and configuration details remain unknown. Source groups are not verified rankings.

GLM-5.2 Risk Evaluation Report, Version 1.1 →

Source-specific extraction of SaferAI · GLM-5.2. Findings and measurements retain their original attribution.

Source transcription is not experimental replication. Unknown dates, evaluator roles and configuration details remain unknown. Source groups are not verified rankings.

CAISI Evaluation of DeepSeek V4 Pro →

Source-specific extraction of CAISI · DeepSeek V4 Pro. Findings and measurements retain their original attribution.

Source transcription is not experimental replication. Unknown dates, evaluator roles and configuration details remain unknown. Source groups are not verified rankings.

Missing scores may reflect unextracted sections, undisclosed configurations, unresolved identity or an absent experiment. These are separate research questions; none implies a zero result.

Interpretation and open questions

Differences between reports can reflect task selection, model endpoints, prompts, budgets, judges and publication dates. Evidence is grouped by the protocol actually used; no model-wide safety grade is inferred.

References

  1. [1] OpenAI — introducing gpt 5 4 ; Publication date and availability section; Parameter counts not disclosed in launch article
  2. [2] METR Time Horizon 1.1 — source data ; results.gpt_5_4
  3. [3] Evaluating DeepSeek v4 Pro for Frontier Risks. 2026-05-29. v1.1 · 29 May 2026 ; PDF p. 9; WMDP-Cyber; accuracy; PDF p. 27; DeceptionBench; deception score: single-turn; PDF p. 27; DeceptionBench; deception score: multi-turn w pressure
  4. [4] Early Results on Evaluating GLM 5.2 for Behavioral and Persuasion Risks. 2026-07-03. 3 July 2026 ; #behavioral; Figure 2 / deceptionbench.png; printed label GPT-5.4 / single-turn; #behavioral; Figure 2 / deceptionbench.png; printed label GPT-5.4 / multi-turn with pressure
  5. [5] GLM-5.2 Risk Evaluation Report. 2026-08-02. v1.1 · revised 26 August 2026
  6. [6] CAISI Evaluation of DeepSeek V4 Pro. 2026-05-01. 1 May 2026 ; Figure 3(b), CTF-Archive Diamond, legend GPT-5.4 (xhigh)
  7. [7] CyberGym official leaderboard and machine-data extraction ; /level1/26/subRows/0/score_10; /level1/50/score_10; /level1/54/score_10
  8. [8] ExploitGym official leaderboard and benchmark page ; /results/4/on_target
  9. [9] SEC-bench Pro official leaderboard: selected 260617 and historical 260505 ; /0/results/1/score_modes/headline/score; /0/results/1/score_modes/completed/score; /1/results/1/score_modes/headline/score