Model launches / OpenAI

GPT-6 Astra

Scores at a glance

137 benchmarks · large dot: GPT-6 Astra; small dots: closest models where reported

Artificial Analysis Intelligence Index v5.0 ↗56.85 index

GPT-6 Astra (Max) · Intelligence Index v5.0 ↗

3 more reported results

52.67 index · GPT-6 Astra (Max) · Intelligence Index v4.3.2 · Artificial Analysis ↗

61.2 · GPT-6 Astra · AA Intelligence Index v4.1.1 · historical · GPT-6 Astra launch report ↗

53 · GPT-6 Astra (max) · AA Index · Meta report snapshot · Muse Wonder official evaluation report ↗

AA-Briefcase v1.1 ↗1568.14 Elo

GPT-6 Astra (Max) · AA-Briefcase v1.1 · overall Elo ↗

1 more reported result

1569 Elo · GPT-6 Astra (max, with fallback) · AA-Briefcase v1.1 · AA-Briefcase v1.1 (Elo-score) · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

AutomationBench-AA ↗68.49%

GPT-6 Astra (Max) · AutomationBench-AA · guardrail-gated objective score ↗

2 more reported results

41.6 · GPT-6 Astra (max) · AutomationBench-AA · Muse Wonder official evaluation report ↗

68.5% · GPT-6 Astra (max, with fallback) · AutomationBench-AA · Automation Bench-AA · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

GDP.pdf ↗31.00%

GPT-6 Astra (Max) · GDP.pdf · all-pass ↗

4 more reported results

34.2 · GPT-6 Astra (max) · GDP.pdf · Muse Wonder official evaluation report ↗

34.2% · GPT-6 Astra (max) · GDP.pdf · GDP.pdf score · % · Epoch AI · GDP.pdf ↗

85.5% · GPT-6 Astra (max, with fallback) · GDP.pdf · GDP.pdf w/ python · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

34.2% · GPT-6 Astra (max, with fallback) · GDP.pdf · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

Terminal-Bench 4.0 ↗59.09%

GPT-6 Astra (Max) · Terminal-Bench 4.0 · AA harness ↗

6 more reported results

59.6 % · GPT-6 Astra (Mini-SWE-agent) · Terminal-Bench 4.0 · Score · Vals AI · Terminal-Bench 4.0 ↗

57.9% · GPT-6 Astra · Terminal-Bench 4.0 · GPT-6 Astra launch report ↗

57.9% · GPT-6 Astra · Terminal-Bench 4.0 · Claude Opus 5.5 launch report ↗

58.2% · GPT-6 Astra · Terminal-Bench 4.0 · Gemini 4 Argon model card ↗

59.1 · GPT-6 Astra (max) · Terminal-Bench 4.0-AA · Muse Wonder official evaluation report ↗

58.2% · GPT-6 Astra (max, with fallback) · Terminal-Bench 4.0 · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

Terminal-Bench-Science 0.1 ↗63.33%

GPT-6 Astra (Max) · Terminal-Bench-Science 0.1 · AA harness ↗

5 more reported results

62.86 % · GPT-6 Astra (Mini-SWE-agent) · Terminal-Bench-Science · Score · Vals AI · Terminal-Bench-Science ↗

64.6% · GPT-6 Astra · Terminal-Bench Science 0.1 · GPT-6 Astra launch report ↗

64.6% · GPT-6 Astra · Terminal-Bench-Science 0.1 · Claude Opus 5.5 launch report ↗

68.1% · GPT-6 Astra · Terminal-Bench Science 0.1 · Gemini 4 Argon model card ↗

68.1% · GPT-6 Astra (max, with fallback) · Terminal-Bench-Science 0.1 · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

FrontierCode 2.0 ↗42.37%

GPT-6 Astra (Max) · FrontierCode 2.0 · AA implementation ↗

CritPt v2 · hard subset ↗70.50%

GPT-6 Astra (Max) · CritPt v2 · hard subset · AA implementation ↗

HLE-Diamond ↗63.80%

GPT-6 Astra (Max) · HLE-Diamond · AA implementation ↗

2 more reported results

60.60% · gpt-6-astra · Scale Labs · Score · Scale Labs / SEAL · Scale Labs ↗

65.8% · GPT-6 Astra (max, with fallback) · HLE-Diamond · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

AA-Omniscience v1.1 ↗42.43 index

GPT-6 Astra (Max) · AA-Omniscience v1.1 · error-penalized index ↗

1 more reported result

43.40 index · GPT-6 Astra (Max) · AA-Omniscience · error-penalized score · Artificial Analysis ↗

AA-Omniscience v1.1 · accuracy ↗61.57%

GPT-6 Astra (Max) · AA-Omniscience v1.1 · accuracy ↗

1 more reported result

62.60% · GPT-6 Astra (Max) · AA-Omniscience · accuracy · Artificial Analysis ↗

AA-Omniscience v1.1 · hallucination rate ↗49.78%

GPT-6 Astra (Max) · AA-Omniscience v1.1 · hallucination rate, lower is better ↗

1 more reported result

51.34% · GPT-6 Astra (Max) · AA-Omniscience · hallucination rate, lower is better · Artificial Analysis ↗

GDPval-AA v2.1 ↗1574.46 Elo

GPT-6 Astra (Max) · GDPval-AA v2.1 · Elo ↗

2 more reported results

1542 · GPT-6 Astra · GDPval-AA v2.1 (Elo) · Claude Opus 5.5 launch report ↗

1573 Elo · GPT-6 Astra (max, with fallback) · GDPval-AA v2.1 · GDPval-AA v2.1 (Elo-score) · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

AA-3D ↗1164.13 Elo

GPT-6 Astra (Max) · AA-3D · Elo ↗

AA-AnalystAgent ↗51.25%

GPT-6 Astra (Max) · AA-AnalystAgent · strict score ↗

Harvey LAB-AA v1.1 ↗8.61%

GPT-6 Astra (Max) · Harvey LAB-AA v1.1 · all-pass ↗

1 more reported result

8.6 · GPT-6 Astra (max) · Harvey LAB-AA · Muse Wonder official evaluation report ↗

FrontierCode 1.1 ↗46.98%

GPT-6 Astra (Max) · FrontierCode 1.1 · AA implementation ↗

4 more reported results

64.5%† · GPT-6 Astra · FrontierCode 1.1 Extended · GPT-6 Astra launch report ↗

53.3%† · GPT-6 Astra · FrontierCode 1.1 Main · GPT-6 Astra launch report ↗

53.3% · GPT-6 Astra · FrontierCode 1.1 Main · Claude Opus 5.5 launch report ↗

53.26% · GPT-6 Astra (max) · FrontierCode · Main score · % · Epoch AI · FrontierCode ↗

AA-LCR v1.1 ↗80.67%

GPT-6 Astra (Max) · AA-LCR v1.1 ↗

SciCode ↗56.48%

GPT-6 Astra (Max) · SciCode · AA implementation ↗

5 more reported results

56.48% · GPT-6 Astra (max) · SciCode · Score · % · Epoch AI · SciCode ↗

55.67% · GPT-6 Astra (xhigh) · SciCode · Score · % · Epoch AI · SciCode ↗

55.44% · GPT-6 Astra (high) · SciCode · Score · % · Epoch AI · SciCode ↗

54.17% · GPT-6 Astra (medium) · SciCode · Score · % · Epoch AI · SciCode ↗

54.05% · GPT-6 Astra (low) · SciCode · Score · % · Epoch AI · SciCode ↗

Humanity’s Last Exam ↗54.68%

GPT-6 Astra (Max) · Humanity’s Last Exam · AA implementation ↗

5 more reported results

57.2% · GPT-6 Astra · Humanity’s Last Exam · tools · GPT-6 Astra launch report ↗

57.2% · tools · GPT-6 Astra · Humanity’s Last Exam · tools · Claude Opus 5.5 launch report ↗

54.8% · GPT-6 Astra (unknown thinking) · HLE · Accuracy · % · Epoch AI · HLE ↗

57.2% · GPT-6 Astra (max, with fallback) · Humanity’s Last Exam · HLE-Full w/ tools · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

53.6% · GPT-6 Astra (max, with fallback) · Humanity’s Last Exam · HLE-Full · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

GPQA Diamond ↗96.06%

GPT-6 Astra (Max) · GPQA Diamond · AA implementation ↗

3 more reported results

95.77% · GPT-6 Astra (max) · GPQA diamond · Best score (across scorers) · % · Epoch AI · GPQA diamond ↗

96.0% · GPT-6 Astra · GPQA Diamond · GPT-6 Astra launch report ↗

96.1% · GPT-6 Astra (max, with fallback) · GPQA Diamond · GPQA-Diamond · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

ITBench-AA ↗48.59%

GPT-6 Astra (Max) · ITBench-AA · root-cause score ↗

MMMU-Pro ↗86.88%

GPT-6 Astra (Max) · MMMU-Pro · AA implementation ↗

2 more reported results

87.8% · GPT-6 Astra (max, with fallback) · MMMU-Pro · MMMU-Pro w/ python · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

86.9% · GPT-6 Astra (max, with fallback) · MMMU-Pro · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

MLCR-AA ↗41.67%

GPT-6 Astra (Max) · MLCR-AA · overall ↗

HealthBench ↗65.98%

GPT-6 Astra (Max) · HealthBench · HealthBench Professional · length-adjusted ↗

2 more reported results

63.4% · GPT-6 Astra · HealthBench · HealthBench Professional · length-adjusted · GPT-6 Astra launch report ↗

65.3 · GPT-6 Astra (max) · HealthBench · HealthBench Professional · length-adjusted · Muse Wonder official evaluation report ↗

CritPt v2 ↗83.14%

GPT-6 Astra (Max) · CritPt v2 · code tools ↗

τ³-Banking ↗41.44%

GPT-6 Astra (Max) · τ³-Banking ↗

Terminal-Bench 2.1 ↗88.39%

GPT-6 Astra (Max) · Terminal-Bench 2.1 ↗

CritPt ↗31.71%

GPT-6 Astra (Max) · CritPt · original ↗

5 more reported results

31.71% · GPT-6 Astra (max) · CritPt · Accuracy · % · Epoch AI · CritPt ↗

31.43% · GPT-6 Astra (xhigh) · CritPt · Accuracy · % · Epoch AI · CritPt ↗

29.14% · GPT-6 Astra (medium) · CritPt · Accuracy · % · Epoch AI · CritPt ↗

28.86% · GPT-6 Astra (high) · CritPt · Accuracy · % · Epoch AI · CritPt ↗

26.29% · GPT-6 Astra (low) · CritPt · Accuracy · % · Epoch AI · CritPt ↗

AA-PersonalityBench ↗54.31 percentile

GPT-6 Astra (Max) · AA-PersonalityBench · expressiveness ↗

7 more reported results

67.00 percentile · GPT-6 Astra (Max) · AA-PersonalityBench · organization · Artificial Analysis ↗

21.31 percentile · GPT-6 Astra (Max) · AA-PersonalityBench · informativeness · Artificial Analysis ↗

94.92 percentile · GPT-6 Astra (Max) · AA-PersonalityBench · transparency · Artificial Analysis ↗

72.08 percentile · GPT-6 Astra (Max) · AA-PersonalityBench · warmth · Artificial Analysis ↗

41.62 percentile · GPT-6 Astra (Max) · AA-PersonalityBench · conviction · Artificial Analysis ↗

72.08 percentile · GPT-6 Astra (Max) · AA-PersonalityBench · imagination · Artificial Analysis ↗

13.69 percentile · GPT-6 Astra (Max) · AA-PersonalityBench · slop · Artificial Analysis ↗

AutomationBench · Zapier ↗41.4%

GPT-6 Astra · AutomationBench ↗

2 more reported results

41.4% · GPT-6 Astra · AutomationBench · Claude Opus 5.5 launch report ↗

41.4% · GPT-6 Astra · AutomationBench · Gemini 4 Argon model card ↗

Vals Index ↗65.02 %

GPT-6 Astra · Vals Index · Accuracy ↗

1 more reported result

63.1% · GPT-6 Astra · Vals Index · Gemini 4 Argon model card ↗

Vals Finance Agent v2 ↗53.54 %

GPT-6 Astra · Finance Agent v2 · Accuracy ↗

3 more reported results

53.5% · GPT-6 Astra · Vals Finance Agent v2 · Gemini 4 Argon model card ↗

53.5 · GPT-6 Astra (max) · Vals Finance Agent v2 · Muse Wonder official evaluation report ↗

45% · GPT-6 Astra (max, with fallback) · Vals Finance Agent v2 · Frontier Finance · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

Harvey Legal Agent Benchmark ↗5.42 %

GPT-6 Astra · Harvey's Legal Agent Benchmark v1 · Accuracy ↗

1 more reported result

5.4% · GPT-6 Astra · Harvey LAB · Vals v1 · Gemini 4 Argon model card ↗

DeepSWE v1.1 ↗74.1%

GPT-6 Astra · DeepSWE v1.1 ↗

8 more reported results

74.1% · GPT-6 Astra · DeepSWE v1.1 · Gemini 4 Argon model card ↗

74.1 · GPT-6 Astra (max) · DeepSWE 1.1 · Muse Wonder official evaluation report ↗

74.12% · GPT-6 Astra (xhigh) · DeepSWE · Pass@1 · % · Epoch AI · DeepSWE ↗

73.23% · GPT-6 Astra (high) · DeepSWE · Pass@1 · % · Epoch AI · DeepSWE ↗

73.23% · GPT-6 Astra (max) · DeepSWE · Pass@1 · % · Epoch AI · DeepSWE ↗

72.79% · GPT-6 Astra (medium) · DeepSWE · Pass@1 · % · Epoch AI · DeepSWE ↗

67.04% · GPT-6 Astra (low) · DeepSWE · Pass@1 · % · Epoch AI · DeepSWE ↗

74.1% · GPT-6 Astra (max, with fallback) · DeepSWE v1.1 · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

FrontierSWE v2 ↗65.5%

GPT-6 Astra · FrontierSWE v2 ↗

1 more reported result

65.51% · GPT-6 Astra (max) · FrontierSWE · Score · % · Epoch AI · FrontierSWE ↗

Vibe Code Bench ↗89.59 %

GPT-6 Astra (OpenHands) · Vibe Code Bench · original task set ↗

2 more reported results

27.64 % · GPT-6 Astra · Vibe Code Bench · Vibe Code Bench 1–100 · Vals AI · Vibe Code Bench 1-100 ↗

89.6% · GPT-6 Astra · Vibe Code Bench · version unspecified in launch report · Gemini 4 Argon model card ↗

PostTrainBench ↗44.3%

GPT-6 Astra · PostTrainBench ↗

1 more reported result

41.9% · GPT-6 Astra (max, with fallback) · PostTrainBench · PostTrain Bench v1.2 · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

LABBench 2 ↗85.4%

GPT-6 Astra · LABBench 2 ↗

RiemannBench ↗72.0%

GPT-6 Astra · RiemannBench ↗

GraphWalks ↗98.7%

GPT-6 Astra · GraphWalks BFS ≤128k · F1 ↗

1 more reported result

71.8% · GPT-6 Astra · GraphWalks BFS 256k–1M · F1 · Gemini 4 Argon model card ↗

Agents’ Last Exam ↗59.3%

GPT-6 Astra · Agents’ Last Exam · source scoring ↗

3 more reported results

34.2% · GPT-6 Astra · Agents’ Last Exam · pass rate · Gemini 4 Argon model card ↗

41.8% · GPT-6 Astra (max, with fallback) · Agents’ Last Exam · Agents' Last Exam (ALL) · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

44.2% · GPT-6 Astra (max, with fallback) · Agents’ Last Exam · Agents' Last Exam (CLI) · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

OSWorld 2.0 ↗72.6%

GPT-6 Astra · OSWorld 2.0 · offline, partial ↗

1 more reported result

72.6% · GPT-6 Astra · OSWorld 2.0 · offline, partial · Gemini 4 Argon model card ↗

Chartography ↗71.0%

GPT-6 Astra · Chartography · no tools ↗

1 more reported result

91% · GPT-6 Astra (max, with fallback) · Chartography · ChartoGraphy w/ python · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

LVBench ↗87.5%

GPT-6 Astra · LVBench ↗

CWE-Bench ↗68.0%

GPT-6 Astra · CWE-Bench v1 ↗

ScreenSpot-Pro ↗92.7%

GPT-6 Astra · ScreenSpot-Pro · no tools ↗

BenchCAD ↗95.9%

GPT-6 Astra · BenchCAD · tools ↗

1 more reported result

95.9% · GPT-6 Astra (max, with fallback) · BenchCAD · BenchCAD w/ tools · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

BrowseComp ↗91.5%

GPT-6 Astra · BrowseComp ↗

1 more reported result

91.5% · GPT-6 Astra (max, with fallback) · BrowseComp · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

FrontierMath Tier 4 ↗97.6%

GPT-6 Astra · FrontierMath Tier 4 · v2 ↗

GeneBench Pro ↗37.1%

GPT-6 Astra · GeneBench Pro ↗

MedChemBench ↗49.3%

GPT-6 Astra · MedChemBench · internal ↗

LifeSciBench ↗60.3%

GPT-6 Astra · LifeSciBench ↗

ExploitBench ↗100.0%

GPT-6 Astra · ExploitBench · original ↗

2 more reported results

39.0% · GPT-6 Astra · ExploitBench · June–August 2026 · GPT-6 Astra launch report ↗

100% · GPT-6 Astra (max, with fallback) · ExploitBench · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

ExploitGym ↗42.4%†

GPT-6 Astra · ExploitGym ↗

1 more reported result

42.4% · GPT-6 Astra (max, with fallback) · ExploitGym · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

SRE-Bench ↗56.87 %

GPT-6 Astra · SRE Bench · Accuracy ↗

1 more reported result

88.0% · GPT-6 Astra · SRE-Bench · GPT-6 Astra launch report ↗

SEC-Bench Pro ↗85.4%

GPT-6 Astra · SEC-Bench Pro ↗

MRCR v2 ↗100.0%

GPT-6 Astra · MRCR v2 · 8 needles, 256k–512k ↗

1 more reported result

96.3% · GPT-6 Astra · MRCR v2 · 8 needles, 512k–1M · GPT-6 Astra launch report ↗

ARC-AGI-3 ↗99.9%†

GPT-6 Astra · ARC-AGI-3 ↗

ARC-AGI-2 ↗95.0%

GPT-6 Astra · ARC-AGI-2 ↗

5 more reported results

95% · GPT-6 Astra (max) · ARC-AGI-2 · Score · % · Epoch AI · ARC-AGI-2 ↗

93.33% · GPT-6 Astra (xhigh) · ARC-AGI-2 · Score · % · Epoch AI · ARC-AGI-2 ↗

92.08% · GPT-6 Astra (high) · ARC-AGI-2 · Score · % · Epoch AI · ARC-AGI-2 ↗

92.08% · GPT-6 Astra (medium) · ARC-AGI-2 · Score · % · Epoch AI · ARC-AGI-2 ↗

85.42% · GPT-6 Astra (low) · ARC-AGI-2 · Score · % · Epoch AI · ARC-AGI-2 ↗

ARC-AGI-1 ↗98.5%

GPT-6 Astra · ARC-AGI-1 ↗

5 more reported results

98.5% · GPT-6 Astra (xhigh) · ARC-AGI · Score · % · Epoch AI · ARC-AGI ↗

98.5% · GPT-6 Astra (high) · ARC-AGI · Score · % · Epoch AI · ARC-AGI ↗

97.5% · GPT-6 Astra (max) · ARC-AGI · Score · % · Epoch AI · ARC-AGI ↗

97.5% · GPT-6 Astra (medium) · ARC-AGI · Score · % · Epoch AI · ARC-AGI ↗

96.5% · GPT-6 Astra (low) · ARC-AGI · Score · % · Epoch AI · ARC-AGI ↗

Personal Superintelligence Index ↗54.5

GPT-6 Astra (max) · Personal Superintelligence Index · internal ↗

JobBench ↗66.1

GPT-6 Astra (max) · JobBench ↗

OSWorld 2.1 ↗75.8% partial / 34.3% binary

GPT-6 Astra (max) · OSWorld 2.1 · partial and binary ↗

1 more reported result

75.8% · GPT-6 Astra (max, with fallback) · OSWorld 2.1 · OSWorld V2.1 · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

MemoryBench ↗81.4

GPT-6 Astra (max) · MemoryBench ↗

SWE-Marathon ↗42.5

GPT-6 Astra (max) · SWE-Marathon ↗

1 more reported result

42.5% · GPT-6 Astra (max, with fallback) · SWE-Marathon · SWE Marathon v1.1 · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

ProgramBench ↗5.5 %

GPT-6 Astra · ProgramBench · Accuracy ↗

2 more reported results

52.5 · GPT-6 Astra (max) · ProgramBench · Muse Wonder official evaluation report ↗

50% · GPT-6 Astra (max, with fallback) · ProgramBench · Program Bench · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

STEMBench ↗2.809 / 4

GPT-6 Astra xhigh · STEMBench120 · mean progress / 4 · tools ↗

ErdősBench ↗3.230 / 4

GPT-6 Astra xhigh · ErdősBench · mean progress / 4 · Codex ↗

FrontierMath Tiers 1-3 (v2) ↗93.68%

GPT-6 Astra (max) · FrontierMath-Tiers-1-3-v2-Private · Best score (across scorers) · % ↗

FrontierMath Tier 4 (v2) ↗97.6%

GPT-6 Astra (high) · FrontierMath-Tier-4-v2-Private · Best score (across scorers) · % ↗

4 more reported results

97.6% · GPT-6 Astra (xhigh) · FrontierMath-Tier-4-v2-Private · Best score (across scorers) · % · Epoch AI · FrontierMath-Tier-4-v2-Private ↗

97.6% · GPT-6 Astra (max) · FrontierMath-Tier-4-v2-Private · Best score (across scorers) · % · Epoch AI · FrontierMath-Tier-4-v2-Private ↗

97.56% · GPT-6 Astra (medium) · FrontierMath-Tier-4-v2-Private · Best score (across scorers) · % · Epoch AI · FrontierMath-Tier-4-v2-Private ↗

87.8% · GPT-6 Astra (low) · FrontierMath-Tier-4-v2-Private · Best score (across scorers) · % · Epoch AI · FrontierMath-Tier-4-v2-Private ↗

OTIS Mock AIME 2024-2025 ↗100%

GPT-6 Astra (max) · OTIS Mock AIME 2024-2025 · Best score (across scorers) · % ↗

SimpleQA Verified ↗75.6%

GPT-6 Astra (max) · SimpleQA Verified · Best score (across scorers) · % ↗

Chess Puzzles ↗76%

GPT-6 Astra (pro, max) · Chess Puzzles · Best score (across scorers) · % ↗

1 more reported result

74% · GPT-6 Astra (max) · Chess Puzzles · Best score (across scorers) · % · Epoch AI · Chess Puzzles ↗

Earthborne Rangers (EBR-bench) ↗76.19%

GPT-6 Astra (max) · EBR-bench · Best score (across scorers) · % ↗

Mystery Game Puzzles ↗84%

GPT-6 Astra (max) · Mystery Game Puzzles · Best score (across scorers) · % ↗

MirrorCode ↗46.7%

GPT-6 Astra (high) · MirrorCode · Best score (across scorers) · % ↗

BALROG ↗68.3%

GPT-6 Astra (max) · Balrog · Average progress · % ↗

GSO ↗79.41%

GPT-6 Astra (unknown thinking) · GSO-Bench · Score OPT@1 · % ↗

WeirdML (v2) ↗93.57%

GPT-6 Astra (pro, max) · WeirdML · Accuracy · % ↗

2 more reported results

93.28% · GPT-6 Astra (max) · WeirdML · Accuracy · % · Epoch AI · WeirdML ↗

92.87% · GPT-6 Astra (high) · WeirdML · Accuracy · % · Epoch AI · WeirdML ↗

APEX-Agents · original ↗64.7%

GPT-6 Astra (unknown thinking) · APEX-Agents · Pass@1 score · % ↗

1 more reported result

64.7% · GPT-6 Astra (max, with fallback) · APEX-Agents · original · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

Remote Labor Index ↗20.83%

GPT 6 Astra · Scale Labs · Score ↗

1 more reported result

20.83% · GPT-6 Astra (unknown thinking) · Remote Labor Index · Score · % · Epoch AI · Remote Labor Index ↗

ProofBench ↗99 %

GPT-6 Astra · ProofBench · Accuracy ↗

1 more reported result

99% · GPT-6 Astra (unknown thinking) · ProofBench · Accuracy · % · Epoch AI · ProofBench ↗

LMCA ↗64.37%

GPT-6 Astra (xhigh) · LMCA · Score · % ↗

4 more reported results

64.11% · GPT-6 Astra (unknown thinking) · LMCA · Score · % · Epoch AI · LMCA ↗

63.84% · GPT-6 Astra (high) · LMCA · Score · % · Epoch AI · LMCA ↗

62.93% · GPT-6 Astra (medium) · LMCA · Score · % · Epoch AI · LMCA ↗

62.19% · GPT-6 Astra (low) · LMCA · Score · % · Epoch AI · LMCA ↗

DTBench ↗97.33%

GPT-6 Astra (xhigh) · DTBench · Accuracy · % ↗

4 more reported results

97.07% · GPT-6 Astra (unknown thinking) · DTBench · Accuracy · % · Epoch AI · DTBench ↗

97.07% · GPT-6 Astra (medium) · DTBench · Accuracy · % · Epoch AI · DTBench ↗

96.53% · GPT-6 Astra (high) · DTBench · Accuracy · % · Epoch AI · DTBench ↗

94.93% · GPT-6 Astra (low) · DTBench · Accuracy · % · Epoch AI · DTBench ↗

FrontierMath: Erdős (Lean 4) ↗2.94%

GPT-6 Astra (max) · FrontierMath-Erdos · Best score (across scorers) · % ↗

Furniture Assembly Benchmark ↗80%

GPT-6 Astra (max) · Furniture Assembly · Best score (across scorers) · % ↗

Text Arena (Coding) ↗1800.28 Elo

GPT-6 Astra (max) · WebDev Arena · Elo ↗

Blueprint-Bench 2 ↗49.67%

GPT-6 Astra (unknown thinking) · Blueprint-Bench 2 · Score · % ↗

WeirdML (v3) ↗54.05%

GPT-6 Astra (xhigh) · WeirdML v3 · Max performance · % ↗

SpireClimb ↗25.67%

GPT-6 Astra (max) · SpireClimb · Best score (across scorers) · % ↗

Epoch Capabilities Index (ECI) ↗168.81 ECI

GPT-6 Astra · General ECI · 26 October 2026 fit ↗

Vals RSI Index ↗0.43 reference score

GPT-6 Astra (max) · Compression · reference-anchored score ↗

3 more reported results

0.48 reference score · GPT-6 Astra (max) · LM training · reference-anchored score · Vals RSI Index · LM training ↗

0.10 reference score · GPT-6 Astra (max) · Harness engineering · reference-anchored score · Vals RSI Index · Harness engineering ↗

0.08 reference score · GPT-6 Astra (max) · Post-training · reference-anchored score · Vals RSI Index · Post-training ↗

Vals Finance Index ↗55.64 %

GPT-6 Astra · Vals Finance Index · Accuracy ↗

Vals Legal Index ↗68.39 %

GPT-6 Astra · Vals Legal Index · Accuracy ↗

Vals Science Index ↗65.09 %

GPT-6 Astra · Vals Science Index · Accuracy ↗

Legal Research Bench ↗39.42 %

GPT-6 Astra · Legal Research Bench · All-Pass ↗

HLAB-Vals-1.1 ↗7.08 %

GPT-6 Astra · HLAB-Vals-1.1 · Task Pass Rate ↗

Vals Finance Agent v2.1 ↗60.43 %

GPT-6 Astra · Finance Agent · Accuracy ↗

Excel Modeling Benchmark (EMB & EMB v1.1) ↗72.68 %

GPT-6 Astra · Excel Modeling Benchmark · Accuracy ↗

1 more reported result

71.7 % · GPT-6 Astra · Excel Modeling Benchmark v1 · Accuracy · Vals AI · Excel Modeling Benchmark v1 ↗

Tax Agent Bench ↗20.72 %

GPT-6 Astra · Tax Agent Bench · Accuracy ↗

MedCode ↗48.49 %

GPT-6 Astra · MedCode · Accuracy ↗

MedScribe ↗87.91 %

GPT-6 Astra · MedScribe · Accuracy ↗

BioMysteryBench ↗79.26 %

GPT-6 Astra · BioMysteryBench · Accuracy ↗

1 more reported result

78.9% · GPT-6 Astra (max, with fallback) · BioMysteryBench · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

MysteryMechanism ↗53.15 %

GPT-6 Astra · MysteryMechanism · Accuracy ↗

Robo-bench ↗23.33 %

GPT-6 Astra · Robo-bench · Accuracy ↗

SAGE (Student Assessment with Generative Evaluation) ↗46.37 %

GPT-6 Astra · SAGE · Accuracy ↗

Code Migration v1.1 ↗80.58 %

GPT-6 Astra · Code Migration · Accuracy ↗

1 more reported result

67.74 % · GPT-6 Astra · Code Migration v1 · Accuracy · Vals AI · Code Migration v1 ↗

IOI v1 & IOI v2 ↗100 %

GPT-6 Astra · IOI · Accuracy ↗

SWE-bench Multimodal ↗43.33 %

GPT-6 Astra · SWE-bench Multimodal · Accuracy ↗

CyberBench v1.1 ↗41.07 %

GPT-6 Astra · CyberBench · Accuracy ↗

CUA-bench ↗19.17 %

GPT-6 Astra · CUA-bench · Accuracy ↗

Time Horizon Index: KSP ↗90.5 %

GPT-6 Astra · Time Horizon Index: KSP · Overall ↗

Humanity's Sixth Sense (HSS) ↗53.60%

GPT-6-Astra (max) · Humanity's Sixth Sense · Score ↗

DistressBench ↗74.31%

gpt-6-astra · DistressBench · Score ↗

DrugDiscoveryBench ↗68.70%

GPT 6 Astra (mini-SWE-agent) max · DrugDiscoveryBench · Score ↗

1 more reported result

65.40% · GPT 6 Astra (Codex) max · DrugDiscoveryBench · Score · Scale Labs / SEAL · DrugDiscoveryBench ↗

SWE Atlas — Refactoring ↗59.05%

GPT 6 Astra (Codex) xHigh · SWE Atlas - Refactoring · Score ↗

SWE Atlas — Test Writing ↗50.74%

GPT 6 Astra (Codex) xHigh · SWE Atlas - Test Writing · Score ↗

HiL-Bench (Human-in-Loop Benchmark) ↗65.32%

GPT 6 Astra · HiL-Bench (Human-in-Loop Benchmark) · Score ↗

SWE-Bench Pro V2 (Full & Hard) ↗96.90%

GPT-6-Astra (Codex) high · SWE-Bench Pro V2 · full ↗

1 more reported result

90.20% · GPT-6 Astra (Codex) high · SWE-Bench Pro V2 · hard · Scale Labs / SEAL · SWE-Bench Pro V2 ↗

PRBench — Finance (Professional Reasoning) ↗47.54%

GPT 6 Astra · Scale Labs · Score ↗

PRBench — Legal (Professional Reasoning) ↗48.36%

GPT 6 Astra · Scale Labs · Score ↗

TasteVal ↗1.6 ×

GPT-6 Astra (max) · TasteVal · Experimental compute multiplier · October 2026 paper comparison ↗

1 more reported result

1.13 × · GPT-6 Astra (max) · TasteVal · Normalized final-performance multiplier · October 2026 paper comparison · TasteVal: Measuring the Experimental Research Taste of AI Systems Against Human Experts ↗

BoxBench ↗77%

GPT-6 Astra · BoxBench · weighted rubric accuracy · OpenAI · ChatGPT financial services launch ↗

FocusBench ↗47.7%

GPT-6 Astra (max) · FocusBench · score · Meta · Muse Wonder launch · maximum effort ↗

Kimi Code Bench 3.0 ↗66.5%

GPT-6 Astra (max, with fallback) · Kimi Code Bench 3.0 · Kimi Code Bench 3.0 Long-horizon (Internal) · Kimi K3.1 Pro · complete release comparison ↗

CharXiv ↗92.8%

GPT-6 Astra (max, with fallback) · CharXiv · CharXiv (RQ) w/ python · Kimi K3.1 Pro · complete release comparison ↗

1 more reported result

90% · GPT-6 Astra (max, with fallback) · CharXiv · CharXiv (RQ) · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

BabyVision ↗96.6%

GPT-6 Astra (max, with fallback) · BabyVision · BabyVision w/ python · Kimi K3.1 Pro · complete release comparison ↗

ZeroBench ↗58%

GPT-6 Astra (max, with fallback) · ZeroBench · ZeroBench_main w/ python (pass@5) · Kimi K3.1 Pro · complete release comparison ↗

2 more reported results

43.2% · GPT-6 Astra (max, with fallback) · ZeroBench · ZeroBench_main · avg@5 · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

52% · GPT-6 Astra (max, with fallback) · ZeroBench · ZeroBench_main · pass@5 · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

OfficeQA Pro ↗54.4%

GPT-6 Astra (max, with fallback) · OfficeQA Pro · OfficeQA Pro v2 ↗

1 more reported result

66.2% · GPT-6 Astra (max, with fallback) · OfficeQA Pro · OfficeQA Pro v1 · Introducing Kimi K3.1 Pro ↗

DECK-Arena ↗1423 Elo

GPT-6 Astra (max, with fallback) · DECK-Arena · DECK-Arena (Internal, Elo-score) · Kimi K3.1 Pro · complete release comparison ↗

Design Bench ↗1076 Elo

GPT-6 Astra (max, with fallback) · Design Bench · Design Bench (Internal, Elo-score) · Kimi K3.1 Pro · complete release comparison ↗

Kimi Hallucination Benchmark ↗5.5%

GPT-6 Astra (max, with fallback) · Kimi Hallucination Benchmark · Kimi Hallu Bench Hard w/ tools (Internal, lower is better) · Kimi K3.1 Pro · complete release comparison ↗

1 more reported result

4.2% · GPT-6 Astra (max, with fallback) · Kimi Hallucination Benchmark · Kimi Hallu Bench w/o tools (Internal, lower is better) · Kimi K3.1 Pro · complete release comparison · Introducing Kimi K3.1 Pro ↗

FrontierMath: Open Problems ↗1 credited problems

GPT-6 Astra · FrontierMath: Open Problems · FM Open Problems · first credited AI solution records · Epoch public solution registry · October 27, 2026 ↗

Vending-Bench 2 ↗15514.70167 USD

GPT-6 Astra · End-of-year balance ↗

BTF-3 ↗0.135 Brier scale

GPT-6 Astra (low) · Pooled forecasting score ↗

GDPx v1.0 ↗9.6%

GPT-6 Astra · GDPx v1.0 · Overall · intended grading ↗

53 more reported results

11.8% · GPT-6 Astra · GDPx v1.0 · Eight-domain subset · intended grading · GDPx v1.0 — SemiAnalysis ↗

24.9% · GPT-6 Astra · GDPx v1.0 · Recursive Self-Improvement · intended grading · GDPx v1.0 — SemiAnalysis ↗

4.4% · GPT-6 Astra · GDPx v1.0 · Chip Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · GPT-6 Astra · GDPx v1.0 · Cybersecurity · intended grading · GDPx v1.0 — SemiAnalysis ↗

5.1% · GPT-6 Astra · GDPx v1.0 · Biology · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · GPT-6 Astra · GDPx v1.0 · Finance · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · GPT-6 Astra · GDPx v1.0 · Legal · intended grading · GDPx v1.0 — SemiAnalysis ↗

20.2% · GPT-6 Astra · GDPx v1.0 · Quant Trading · intended grading · GDPx v1.0 — SemiAnalysis ↗

16.5% · GPT-6 Astra · GDPx v1.0 · Hardware Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

21.6% · GPT-6 Astra · GDPx v1.0 · Materials Science · intended grading · GDPx v1.0 — SemiAnalysis ↗

6.2% · GPT-6 Astra · GDPx v1.0 · Alignment · intended grading · GDPx v1.0 — SemiAnalysis ↗

6.7% · GPT-6 Astra · GDPx v1.0 · Software Engineering · intended grading · GDPx v1.0 — SemiAnalysis ↗

6.2% · GPT-6 Astra · GDPx v1.0 · Aria · Chip Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · GPT-6 Astra · GDPx v1.0 · Arcverex · Chip Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

7.1% · GPT-6 Astra · GDPx v1.0 · Ergodex · Chip Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · GPT-6 Astra · GDPx v1.0 · Arithmetic · Cybersecurity · intended grading · GDPx v1.0 — SemiAnalysis ↗

5.1% · GPT-6 Astra · GDPx v1.0 · LatchBio · Biology · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · GPT-6 Astra · GDPx v1.0 · Living Context · Finance · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · GPT-6 Astra · GDPx v1.0 · Chakra · Legal · intended grading · GDPx v1.0 — SemiAnalysis ↗

20.2% · GPT-6 Astra · GDPx v1.0 · Third Moment · Quant Trading · intended grading · GDPx v1.0 — SemiAnalysis ↗

24.9% · GPT-6 Astra · GDPx v1.0 · Extrapolate · Recursive Self-Improvement · intended grading · GDPx v1.0 — SemiAnalysis ↗

16.5% · GPT-6 Astra · GDPx v1.0 · Beautiful Machines · Hardware Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

26.6% · GPT-6 Astra · GDPx v1.0 · ValenceForge · Materials Science · intended grading · GDPx v1.0 — SemiAnalysis ↗

16.5% · GPT-6 Astra · GDPx v1.0 · Emulated · Materials Science · intended grading · GDPx v1.0 — SemiAnalysis ↗

6.2% · GPT-6 Astra · GDPx v1.0 · Fleet · Alignment · intended grading · GDPx v1.0 — SemiAnalysis ↗

6.7% · GPT-6 Astra · GDPx v1.0 · Datacurve · Software Engineering · intended grading · GDPx v1.0 — SemiAnalysis ↗

19.2% · GPT-6 Astra · GDPx v1.0 · Overall · partial credit · GDPx v1.0 — SemiAnalysis ↗

24.3% · GPT-6 Astra · GDPx v1.0 · Eight-domain subset · partial credit · GDPx v1.0 — SemiAnalysis ↗

24.9% · GPT-6 Astra · GDPx v1.0 · Recursive Self-Improvement · partial credit · GDPx v1.0 — SemiAnalysis ↗

9.7% · GPT-6 Astra · GDPx v1.0 · Chip Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

0.0% · GPT-6 Astra · GDPx v1.0 · Cybersecurity · partial credit · GDPx v1.0 — SemiAnalysis ↗

10.3% · GPT-6 Astra · GDPx v1.0 · Biology · partial credit · GDPx v1.0 — SemiAnalysis ↗

52.9% · GPT-6 Astra · GDPx v1.0 · Finance · partial credit · GDPx v1.0 — SemiAnalysis ↗

31.0% · GPT-6 Astra · GDPx v1.0 · Legal · partial credit · GDPx v1.0 — SemiAnalysis ↗

20.2% · GPT-6 Astra · GDPx v1.0 · Quant Trading · partial credit · GDPx v1.0 — SemiAnalysis ↗

16.5% · GPT-6 Astra · GDPx v1.0 · Hardware Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

32.6% · GPT-6 Astra · GDPx v1.0 · Materials Science · partial credit · GDPx v1.0 — SemiAnalysis ↗

6.2% · GPT-6 Astra · GDPx v1.0 · Alignment · partial credit · GDPx v1.0 — SemiAnalysis ↗

6.7% · GPT-6 Astra · GDPx v1.0 · Software Engineering · partial credit · GDPx v1.0 — SemiAnalysis ↗

22.1% · GPT-6 Astra · GDPx v1.0 · Aria · Chip Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

0.0% · GPT-6 Astra · GDPx v1.0 · Arcverex · Chip Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

7.1% · GPT-6 Astra · GDPx v1.0 · Ergodex · Chip Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

0.0% · GPT-6 Astra · GDPx v1.0 · Arithmetic · Cybersecurity · partial credit · GDPx v1.0 — SemiAnalysis ↗

10.3% · GPT-6 Astra · GDPx v1.0 · LatchBio · Biology · partial credit · GDPx v1.0 — SemiAnalysis ↗

52.9% · GPT-6 Astra · GDPx v1.0 · Living Context · Finance · partial credit · GDPx v1.0 — SemiAnalysis ↗

31.0% · GPT-6 Astra · GDPx v1.0 · Chakra · Legal · partial credit · GDPx v1.0 — SemiAnalysis ↗

20.2% · GPT-6 Astra · GDPx v1.0 · Third Moment · Quant Trading · partial credit · GDPx v1.0 — SemiAnalysis ↗

24.9% · GPT-6 Astra · GDPx v1.0 · Extrapolate · Recursive Self-Improvement · partial credit · GDPx v1.0 — SemiAnalysis ↗

16.5% · GPT-6 Astra · GDPx v1.0 · Beautiful Machines · Hardware Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

44.0% · GPT-6 Astra · GDPx v1.0 · ValenceForge · Materials Science · partial credit · GDPx v1.0 — SemiAnalysis ↗

21.3% · GPT-6 Astra · GDPx v1.0 · Emulated · Materials Science · partial credit · GDPx v1.0 — SemiAnalysis ↗

6.2% · GPT-6 Astra · GDPx v1.0 · Fleet · Alignment · partial credit · GDPx v1.0 — SemiAnalysis ↗

6.7% · GPT-6 Astra · GDPx v1.0 · Datacurve · Software Engineering · partial credit · GDPx v1.0 — SemiAnalysis ↗

Closest models by score

Similar positions on shared benchmark leaderboards. Smaller gaps are closer.

GPT-6.1 Sol →OpenAI

6.9 percentile-point gap

61 of 76 shared benchmark comparisons

Claude Opus 5.5 →Anthropic

17.4 percentile-point gap

66 of 76 shared benchmark comparisons

Claude Sonnet 5.5 →Anthropic

19.3 percentile-point gap

54 of 76 shared benchmark comparisons

How similarity is measured

Mean absolute gap in percentile positions on shared benchmark settings; lower is closer. At least 60% overlap is required, with a minimum of three when available. Independent evaluations are used when there are at least three; otherwise published report comparisons are used. Repeated model observations within one setting are averaged before ranking. Writing-style traits and composite indices are excluded. This describes similarity, not an overall quality ranking.

About the model report

Headline scores can use the best reasoning effort for each evaluation. Internal, public, tool-enabled, and no-tools measurements remain distinct. Reviewed 2026-10-25.

Read the official model report ↗

25 benchmarks in the report

Search this collection →

COD / BENCHMARK115 reported results

Terminal-Bench 4.0

Complex terminal work spans coding, science, infrastructure, and other technical tasks, with executable checks.

Verified task success↗

SCI / BENCHMARK90 reported results

Terminal-Bench-Science 0.1

Research workflows are completed in a terminal and checked with task-specific tests.

Verified task success↗

COD / BENCHMARK95 reported results

FrontierCode 1.1

Repository-level engineering is evaluated against the standard of work maintainers could accept.

Task success↗

RSN / BENCHMARK131 reported results

Humanity’s Last Exam

Expert-authored questions stretch academic reasoning and knowledge across many disciplines.

Answer accuracy↗

SCI / BENCHMARK507 reported results

GPQA Diamond

A demanding subset of expert-written science questions tests detailed reasoning in specialist domains.

Multiple-choice accuracy↗

KNW / BENCHMARK30 reported results

HealthBench

Physician-authored rubrics evaluate health conversations. Original HealthBench and the Professional evaluation use distinct measures and settings.

Rubric score↗

RSN / BENCHMARK279 reported results

ARC-AGI-2

Novel grid-transformation problems test abstraction and adaptation from a small number of examples.

Task success↗

AGT / BENCHMARK12 reported results

BrowseComp

Difficult information-seeking questions test a browsing agent’s ability to locate and combine evidence.

Answer accuracy↗

COD / BENCHMARK97 reported results

DeepSWE v1.1

Original engineering problems test coding agents on substantial changes in existing repositories.

Verified task success↗

AGT / BENCHMARK24 reported results

Agents’ Last Exam

Professional software workflows test whether an agent can deliver the requested result in a real application environment.

Professional task success↗

AGT / BENCHMARK22 reported results

OSWorld 2.0

Long-horizon computer workflows extend evaluation across applications, files, and realistic task sequences.

Partial and binary task score↗

MTH / BENCHMARK5 reported results

FrontierMath Tier 4

The hardest FrontierMath tier tests advanced mathematical problem solving in a particularly demanding slice of the suite.

Hardest-tier correctness↗

CTX / BENCHMARK4 reported results

MRCR v2

Long-context tasks require tracking multiple references and retrieving the requested information from a large input.

Retrieval and reference resolution↗

RSN / BENCHMARK300 reported results

ARC-AGI-1

Few-shot grid transformations test the discovery and application of unfamiliar rules.

Grid-task success↗

RSN / BENCHMARK3 reported results

ARC-AGI-3

Interactive environments assess how an agent explores, learns rules, and adapts to unfamiliar tasks.

Interactive adaptation score↗

AGT / BENCHMARK7 reported results

ScreenSpot-Pro

A model must locate the interface element referred to by a user instruction on a professional screenshot.

Visual grounding accuracy↗

COD / BENCHMARK14 reported results

BenchCAD

Visual-to-code tasks assess whether a model can construct a CAD representation matching a target shape.

Geometric reconstruction overlap↗

SCI / BENCHMARK2 reported results

LifeSciBench

Life-science questions probe specialist reasoning and the use of relevant scientific knowledge.

Life-science task score↗

SCI / BENCHMARK2 reported results

GeneBench Pro

Specialist genetics evaluation appears in recent frontier-model science reports.

Genetics task score↗

SCI / BENCHMARK2 reported results

MedChemBench

OpenAI’s internal medicinal-chemistry evaluation measures a specialist slice of scientific reasoning.

Medicinal-chemistry task score↗

COD / BENCHMARK30 reported results

ExploitBench

A controlled security evaluation measures model capability on vulnerability-related software tasks.

Security capability score↗

COD / BENCHMARK10 reported results

ExploitGym

A controlled security benchmark is used in frontier-model capability and safeguard evaluations.

Security task success↗

COD / BENCHMARK38 reported results

SRE-Bench

In the GPT-6 Astra report, this name refers to software reverse engineering rather than site-reliability incident handling.

Reverse-engineering accuracy↗

COD / BENCHMARK2 reported results

SEC-Bench Pro

Security-oriented software evaluation appears in recent model capability reports.

Security engineering score↗

AGT / BENCHMARK20 reported results

AutomationBench · Zapier

Business workflows require an agent to coordinate real tools and leave the simulated business systems in the required final state.

Verified final-state success↗

COMPARISON CHECKLIST

Match benchmark version, task subset, model effort, harness, tool access, attempt budget, scoring rule, and fallback behavior. A shared benchmark name is the start of a comparison, not proof that two reported scores use the same contract.

OpenAI original report ↗