THE SCOREBOARD FOR WHAT'S NEXTAI + QUANTUM, SCORED. COMPARED. EXPLAINED.
AIQUANTUMSCORE.

Google / Gemini 4

Gemini 4 Argon

Version gemini-4-argon · Released date unverified · Available through Gemini API

Compare this model ↗
Insufficient verified data49% coverage

Explore Google history →

INDEPENDENT BENCHMARK / SNAPSHOT 2026-10-16

Independent benchmark snapshot

This snapshot is over 14 days old. Review the dated benchmark records below for newer evidence.

Artificial Analysis Intelligence Index v4.3.2 · Higher is better. Selected models, sorted by this benchmark’s published score; equal scores are ties.

ModelAA Intelligence IndexTested effortEvidence
Gemini 4 Argon53HighIndependent result ↗

This is a dated snapshot from one independent evaluator, on its own scale—not a Quantum Score or a complete market ranking. A missing Quantum Score means our broader evidence requirements are unmet; it does not mean the model is worse. Test dates are not published in these source summaries.

THE PRACTICAL READ

Should you shortlist Gemini 4 Argon?

Evidence checked 2026-11-07. This interpretation uses the cited records below; it is not a hands-on review.

What the evidence supports

The strongest relative category results in our current evidence are reasoning and agents. Scores describe this comparison pool, not a percentage of real-world tasks solved.

7 current independent benchmark observations. Several tasks from one suite do not count as separate independent sources.

What remains uncertain

Comparable category scores are missing for multimodal. Writing quality, factual accuracy, privacy and production reliability require separate evaluation.

An overall score is withheld because the evidence requirements are not met. Check the effort level and tool configuration before comparing results.

Inspect coverage →

Cost and response time

A current sourced monthly API estimate is unavailable. This excludes tools, caching, batch rates, subscriptions and taxes.

Output tokens per second measures generation after it starts. Reasoning time can make a fast-generating model take longer to finish. Use the measured speed and effort records below.

Estimate your actual usage →

Alternatives worth comparing

Follow changes to this model ↗ Choose an AI for coding → · Research guide → · Automation guide →

Coding48
Reasoning100
MathNo data
REVAL™
ⓘREVAL™ Research Evaluation Index — our proprietary 0–100 benchmark measuring real-world AI research performance. Methodology →
Not tested
Agents73
MultimodalNo data
SpeedNot measured
ValueNo data

PROPRIETARY / RESEARCH EVALUATION INDEX

REVAL™ Not tested

Methodology →

This exact model version has not completed a reviewed REVAL evaluation. No score is assigned from reputation or external benchmarks.

01 / EVIDENCE

Benchmark breakdown

BenchmarkVersionMethodConfigurationResultTest dateVerifiedSource typeSource
Humanity's Last Exam AA October frontier cohortAA Index 4.3.2Artificial Analysis independent evaluation; published reasoning configuration per modelAA Index 4.3.2 October frontier comparison; first-party models; effort recorded separately · High57 %not published2026-10-16independent labView source ↗
AutomationBench-AA AA October frontier cohortAA Index 4.3.2Artificial Analysis independent evaluation; published reasoning configuration per modelAA Index 4.3.2 October frontier comparison; first-party models; effort recorded separately · High78 %not published2026-10-16independent labView source ↗
Terminal-Bench 4.0 AA October frontier cohortAA Index 4.3.2Artificial Analysis independent evaluation; published reasoning configuration per modelAA Index 4.3.2 October frontier comparison; first-party models; effort recorded separately · High57 %not published2026-10-16independent labView source ↗
AA-LCR v1.1 AA October frontier cohortAA Index 4.3.2Artificial Analysis independent evaluation; published reasoning configuration per modelAA Index 4.3.2 October frontier comparison; first-party models; effort recorded separately · High80 %not published2026-10-16independent labView source ↗
LiveBench 2.0 coding20261025LiveBench 2.0 published release tablePublished model configuration; effort not specified in table89.7 %not published2026-11-07benchmark maintainerView source ↗
LiveBench 2.0 reasoning20261025LiveBench 2.0 published release tablePublished model configuration; effort not specified in table95.1 %not published2026-11-07benchmark maintainerView source ↗
LiveBench 2.0 long_agentic20261025LiveBench 2.0 published release tablePublished model configuration; effort not specified in table61.2 %not published2026-11-07benchmark maintainerView source ↗

02 / COST

Price & speed

Input / 1M tokens —

Output / 1M tokens —

Cached input / 1M —

Batch input / 1M —

Batch output / 1M —

Speed No comparable observation yet

Price verified —

02A / PERFORMANCE

Measured speed

No sourced speed measurement at a stated effort level yet.

03 / PROFILE

Capabilities

tool usevisioncode executionstructured outputs

04 / CONTEXT

Evidence coverage & limits

Published tests describe specific environments and effort settings. They do not establish performance on every task. Read the practical interpretation above →

Coverage: 49% · Limited evidence · 7 observations · 2 source domains (2 independent).

Why no Overall AIQuantumScore?

Qualified category evidence: coding, reasoning, agents. An overall score is withheld because only 2 benchmark source domains (minimum 3); only 7 independent observations (minimum 8); only 49% of the required weighted category coverage; only 3 qualified categories (minimum 4). Evidence still missing or lacking comparable overlap: math, research, multimodal, speed.

See the benchmark coverage matrix →

05 / HISTORY

Version & price history

Version gemini-4-argon · released date unverified.

06 / SOURCES

Source records

DATED EVIDENCE

History

Standard API price history

No comparable historical observations yet.

Humanity's Last Exam AA October frontier cohort / AA Index 4.3.2 / %

2026-10-162026-10-16
  • 2026-10-16: 57 result · Artificial Analysis independent evaluation; published reasoning configuration per model · evaluation date not published; verification date shown · source ↗

One observation; no trend can be inferred.

AutomationBench-AA AA October frontier cohort / AA Index 4.3.2 / %

2026-10-162026-10-16
  • 2026-10-16: 78 result · Artificial Analysis independent evaluation; published reasoning configuration per model · evaluation date not published; verification date shown · source ↗

One observation; no trend can be inferred.

Terminal-Bench 4.0 AA October frontier cohort / AA Index 4.3.2 / %

2026-10-162026-10-16
  • 2026-10-16: 57 result · Artificial Analysis independent evaluation; published reasoning configuration per model · evaluation date not published; verification date shown · source ↗

One observation; no trend can be inferred.

AA-LCR v1.1 AA October frontier cohort / AA Index 4.3.2 / %

2026-10-162026-10-16
  • 2026-10-16: 80 result · Artificial Analysis independent evaluation; published reasoning configuration per model · evaluation date not published; verification date shown · source ↗

One observation; no trend can be inferred.

LiveBench 2.0 coding / 20261025 / %

2026-11-072026-11-07
  • 2026-11-07: 89.7 result · LiveBench 2.0 published release table · evaluation date not published; verification date shown · source ↗

One observation; no trend can be inferred.

LiveBench 2.0 reasoning / 20261025 / %

2026-11-072026-11-07
  • 2026-11-07: 95.1 result · LiveBench 2.0 published release table · evaluation date not published; verification date shown · source ↗

One observation; no trend can be inferred.

LiveBench 2.0 long_agentic / 20261025 / %

2026-11-072026-11-07
  • 2026-11-07: 61.2 result · LiveBench 2.0 published release table · evaluation date not published; verification date shown · source ↗

One observation; no trend can be inferred.

Category score history

No two comparable ranking snapshots have been recorded yet.

07 / EXPLORE

Similar models