RLTasksby ulam.ai
Model launches / Google

Gemini 4 Argon

Scores at a glance

98 benchmarks · large dot: Gemini 4 Argon; small dots: closest models where reported

AA-Omniscience v1.1 ↗46.10 index
-8.2748.13

Gemini 4 Argon (High) · AA-Omniscience v1.1 · error-penalized index ↗

1 more reported result

42.35 index · Gemini 4 Argon (High) · AA-Omniscience · error-penalized score · Artificial Analysis ↗

AA-3D ↗1152.64 Elo
572.551198.83

Gemini 4 Argon (High) · AA-3D · Elo ↗

SciCode ↗61.81%
42.9466.9

Gemini 4 Argon (High) · SciCode · AA implementation ↗

1 more reported result

61.81% · Gemini 4 Argon (high) · SciCode · Score · % · Epoch AI · SciCode ↗

MLCR-AA ↗13.89%
6.1141.67

Gemini 4 Argon (High) · MLCR-AA · overall ↗

HealthBench ↗55.05%
50.765.98

Gemini 4 Argon (High) · HealthBench · HealthBench Professional · length-adjusted ↗

CritPt ↗27.14%
4.5732.86

Gemini 4 Argon (High) · CritPt · original ↗

1 more reported result

27.14% · Gemini 4 Argon (high) · CritPt · Accuracy · % · Epoch AI · CritPt ↗

AA-PersonalityBench ↗82.23 percentile
28.92100

Gemini 4 Argon (High) · AA-PersonalityBench · expressiveness ↗

7 more reported results

51.77 percentile · Gemini 4 Argon (High) · AA-PersonalityBench · organization · Artificial Analysis ↗

79.69 percentile · Gemini 4 Argon (High) · AA-PersonalityBench · informativeness · Artificial Analysis ↗

16.23 percentile · Gemini 4 Argon (High) · AA-PersonalityBench · transparency · Artificial Analysis ↗

74.62 percentile · Gemini 4 Argon (High) · AA-PersonalityBench · warmth · Artificial Analysis ↗

59.38 percentile · Gemini 4 Argon (High) · AA-PersonalityBench · conviction · Artificial Analysis ↗

44.15 percentile · Gemini 4 Argon (High) · AA-PersonalityBench · imagination · Artificial Analysis ↗

69.54 percentile · Gemini 4 Argon (High) · AA-PersonalityBench · slop · Artificial Analysis ↗

STEMBench ↗2.444 / 4
2.172.68

Gemini 4 Argon · STEMBench120 · mean progress / 4 · closed ↗

ErdősBench ↗3.040 / 4
2.113.25

Gemini 4 Argon · ErdősBench · mean progress / 4 · API ↗

GPQA Diamond ↗96.4%
9.2896.4

Gemini 4 Argon (high) · GPQA diamond · Best score (across scorers) · % ↗

Chess Puzzles ↗59%
076

Gemini 4 Argon (high) · Chess Puzzles · Best score (across scorers) · % ↗

MirrorCode ↗58.89%
8.8977.4

Gemini 4 Argon (high) · MirrorCode · Best score (across scorers) · % ↗

Vals RSI Index ↗0.45 reference score
00.46

Gemini 4 Argon (high) · Compression · reference-anchored score ↗

3 more reported results

0.48 reference score · Gemini 4 Argon (high) · LM training · reference-anchored score · Vals RSI Index · LM training ↗

0.14 reference score · Gemini 4 Argon (high) · Harness engineering · reference-anchored score · Vals RSI Index · Harness engineering ↗

0.15 reference score · Gemini 4 Argon (high) · Post-training · reference-anchored score · Vals RSI Index · Post-training ↗

SAFE-Teen ↗53%
1556

Gemini 4 Argon · Urgent safety · share of applicable ratings marked fail ↗

2 more reported results

71% · Gemini 4 Argon · Health advice · share of applicable ratings marked fail · Vals AI · SAFE-Teen ↗

66% · Gemini 4 Argon · Relationship boundaries · share of applicable ratings marked fail · Vals AI · SAFE-Teen ↗

GDPx v1.0 ↗7.3%
1.6712.03

Gemini 4 Argon · GDPx v1.0 · Overall · intended grading ↗

53 more reported results

5.4% · Gemini 4 Argon · GDPx v1.0 · Eight-domain subset · intended grading · GDPx v1.0 — SemiAnalysis ↗

14.1% · Gemini 4 Argon · GDPx v1.0 · Recursive Self-Improvement · intended grading · GDPx v1.0 — SemiAnalysis ↗

1.6% · Gemini 4 Argon · GDPx v1.0 · Chip Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

18.5% · Gemini 4 Argon · GDPx v1.0 · Cybersecurity · intended grading · GDPx v1.0 — SemiAnalysis ↗

12.8% · Gemini 4 Argon · GDPx v1.0 · Biology · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · Gemini 4 Argon · GDPx v1.0 · Finance · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · Gemini 4 Argon · GDPx v1.0 · Legal · intended grading · GDPx v1.0 — SemiAnalysis ↗

6.2% · Gemini 4 Argon · GDPx v1.0 · Quant Trading · intended grading · GDPx v1.0 — SemiAnalysis ↗

2.7% · Gemini 4 Argon · GDPx v1.0 · Hardware Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

10.2% · Gemini 4 Argon · GDPx v1.0 · Materials Science · intended grading · GDPx v1.0 — SemiAnalysis ↗

5.8% · Gemini 4 Argon · GDPx v1.0 · Alignment · intended grading · GDPx v1.0 — SemiAnalysis ↗

8.9% · Gemini 4 Argon · GDPx v1.0 · Software Engineering · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.6% · Gemini 4 Argon · GDPx v1.0 · Aria · Chip Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · Gemini 4 Argon · GDPx v1.0 · Arcverex · Chip Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

4.0% · Gemini 4 Argon · GDPx v1.0 · Ergodex · Chip Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

18.5% · Gemini 4 Argon · GDPx v1.0 · Arithmetic · Cybersecurity · intended grading · GDPx v1.0 — SemiAnalysis ↗

12.8% · Gemini 4 Argon · GDPx v1.0 · LatchBio · Biology · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · Gemini 4 Argon · GDPx v1.0 · Living Context · Finance · intended grading · GDPx v1.0 — SemiAnalysis ↗

0.0% · Gemini 4 Argon · GDPx v1.0 · Chakra · Legal · intended grading · GDPx v1.0 — SemiAnalysis ↗

6.2% · Gemini 4 Argon · GDPx v1.0 · Third Moment · Quant Trading · intended grading · GDPx v1.0 — SemiAnalysis ↗

14.1% · Gemini 4 Argon · GDPx v1.0 · Extrapolate · Recursive Self-Improvement · intended grading · GDPx v1.0 — SemiAnalysis ↗

2.7% · Gemini 4 Argon · GDPx v1.0 · Beautiful Machines · Hardware Design · intended grading · GDPx v1.0 — SemiAnalysis ↗

9.9% · Gemini 4 Argon · GDPx v1.0 · ValenceForge · Materials Science · intended grading · GDPx v1.0 — SemiAnalysis ↗

10.4% · Gemini 4 Argon · GDPx v1.0 · Emulated · Materials Science · intended grading · GDPx v1.0 — SemiAnalysis ↗

5.8% · Gemini 4 Argon · GDPx v1.0 · Fleet · Alignment · intended grading · GDPx v1.0 — SemiAnalysis ↗

8.9% · Gemini 4 Argon · GDPx v1.0 · Datacurve · Software Engineering · intended grading · GDPx v1.0 — SemiAnalysis ↗

13.5% · Gemini 4 Argon · GDPx v1.0 · Overall · partial credit · GDPx v1.0 — SemiAnalysis ↗

12.7% · Gemini 4 Argon · GDPx v1.0 · Eight-domain subset · partial credit · GDPx v1.0 — SemiAnalysis ↗

14.1% · Gemini 4 Argon · GDPx v1.0 · Recursive Self-Improvement · partial credit · GDPx v1.0 — SemiAnalysis ↗

3.9% · Gemini 4 Argon · GDPx v1.0 · Chip Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

18.5% · Gemini 4 Argon · GDPx v1.0 · Cybersecurity · partial credit · GDPx v1.0 — SemiAnalysis ↗

22.2% · Gemini 4 Argon · GDPx v1.0 · Biology · partial credit · GDPx v1.0 — SemiAnalysis ↗

19.9% · Gemini 4 Argon · GDPx v1.0 · Finance · partial credit · GDPx v1.0 — SemiAnalysis ↗

26.0% · Gemini 4 Argon · GDPx v1.0 · Legal · partial credit · GDPx v1.0 — SemiAnalysis ↗

6.2% · Gemini 4 Argon · GDPx v1.0 · Quant Trading · partial credit · GDPx v1.0 — SemiAnalysis ↗

2.7% · Gemini 4 Argon · GDPx v1.0 · Hardware Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

20.1% · Gemini 4 Argon · GDPx v1.0 · Materials Science · partial credit · GDPx v1.0 — SemiAnalysis ↗

5.8% · Gemini 4 Argon · GDPx v1.0 · Alignment · partial credit · GDPx v1.0 — SemiAnalysis ↗

8.9% · Gemini 4 Argon · GDPx v1.0 · Software Engineering · partial credit · GDPx v1.0 — SemiAnalysis ↗

7.8% · Gemini 4 Argon · GDPx v1.0 · Aria · Chip Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

0.0% · Gemini 4 Argon · GDPx v1.0 · Arcverex · Chip Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

4.0% · Gemini 4 Argon · GDPx v1.0 · Ergodex · Chip Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

18.5% · Gemini 4 Argon · GDPx v1.0 · Arithmetic · Cybersecurity · partial credit · GDPx v1.0 — SemiAnalysis ↗

22.2% · Gemini 4 Argon · GDPx v1.0 · LatchBio · Biology · partial credit · GDPx v1.0 — SemiAnalysis ↗

19.9% · Gemini 4 Argon · GDPx v1.0 · Living Context · Finance · partial credit · GDPx v1.0 — SemiAnalysis ↗

26.0% · Gemini 4 Argon · GDPx v1.0 · Chakra · Legal · partial credit · GDPx v1.0 — SemiAnalysis ↗

6.2% · Gemini 4 Argon · GDPx v1.0 · Third Moment · Quant Trading · partial credit · GDPx v1.0 — SemiAnalysis ↗

14.1% · Gemini 4 Argon · GDPx v1.0 · Extrapolate · Recursive Self-Improvement · partial credit · GDPx v1.0 — SemiAnalysis ↗

2.7% · Gemini 4 Argon · GDPx v1.0 · Beautiful Machines · Hardware Design · partial credit · GDPx v1.0 — SemiAnalysis ↗

25.2% · Gemini 4 Argon · GDPx v1.0 · ValenceForge · Materials Science · partial credit · GDPx v1.0 — SemiAnalysis ↗

15.0% · Gemini 4 Argon · GDPx v1.0 · Emulated · Materials Science · partial credit · GDPx v1.0 — SemiAnalysis ↗

5.8% · Gemini 4 Argon · GDPx v1.0 · Fleet · Alignment · partial credit · GDPx v1.0 — SemiAnalysis ↗

8.9% · Gemini 4 Argon · 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.

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

The model card and methodology separate self-computed results from external leaderboards, and disclose hardware, frame budgets, task subsets, and scoring choices. Reviewed 2026-10-25.

Read the official model report ↗

18 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 / BENCHMARK97 reported results

DeepSWE v1.1

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

Verified task success↗
RSN / BENCHMARK22 reported results

PostTrainBench

An agent must improve small language models within a constrained training-compute budget.

Held-out post-training performance↗
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↗
RSN / BENCHMARK17 reported results

Chartography

Specialist charts test visual reasoning grounded in professional practice.

Professional chart accuracy↗
COD / BENCHMARK25 reported results

FrontierSWE v2

Extended engineering and research challenges test how much useful work an agent can complete within a long fixed budget.

Mean task score across trials↗
COD / BENCHMARK149 reported results

Vibe Code Bench

Compare app-building outcomes across the original suite and the 1–100 collection, with each task set labeled separately.

Application-building score↗
SCI / BENCHMARK4 reported results

LABBench 2

Scientific tasks evaluate reasoning and computational work with domain-relevant tools.

Scientific workflow score↗
MTH / BENCHMARK4 reported results

RiemannBench

A mathematical evaluation is included in recent frontier-model science and math comparisons.

Mathematical reasoning score↗
CTX / BENCHMARK8 reported results

GraphWalks

A long input describes a graph, and the model must retrieve related nodes or follow graph structure to the requested depth.

Node-set F1↗
RSN / BENCHMARK10 reported results

LVBench

A language model with visual input must understand and reason over long video material.

Long-video understanding accuracy↗
COD / BENCHMARK4 reported results

CWE-Bench

A controlled software-security evaluation contributes to recent frontier-model capability comparisons.

Controlled security task score↗
KNW / BENCHMARK31 reported results

Vals Index

A composite view summarizes model performance across professional-domain evaluations.

Professional capability composite↗
FIN / BENCHMARK96 reported results

Vals Finance Agent v2

Agentic financial research evaluates whether a system can use tools and evidence to complete analyst work.

Financial research task 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↗
KNW / BENCHMARK88 reported results

Harvey Legal Agent Benchmark

A client matter provides documents and a legal-work request; the agent must deliver a reviewable artifact satisfying expert-written requirements.

Expert-rubric all-pass rate↗

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.

Google original report ↗