readable AI benchmarks (simplified)

readable AI benchmarks

Google

Gemini 4 Argon (High)

reasoning · closed weights · Sep 30, 2026

Gemini 4 Argon (High) has a Quality Score of 61.7, ranking 27th among 283 scored models. Per 1M tokens, pricing is $2.00 input, $10.00 output, and $0.100 cached input. It accepts text, image, video, audio, and pdf input, outputs text, and has a 1M token context window. Its Reliability Score is 73.0, ranking 17th among 288 scored models, above average (total average: 48.2). Its Value Score is 41.2, high compared with the total Value average of 29.3.

Quality Score61.7coverage 100.0%: AA Intelligence and AA-Omniscience outcomes available
Value Score41.2
Factual reliability73.0
Cache Discount95%

Model specification

Reasoning
reasoning
Input modalities
text, image, video, audio, pdf
Output modalities
text
Context window
1M tokens
Weights
closed weights

Token prices USD per 1M tokens

Input
$2.00
Output
$10.00
Cached Input
$0.100
Cache write
not available
Cost per task
$4.97

Capability

Intelligence
50.4
Omniscience (native net)
46.1
Correct
48.3%
Blended price
$1.47

Answer outcomes

Fully graded outcomes (Correct + Incorrect): 50.5%

Correct 48.3%Incorrect 2.2%Partial / not attempted 49.5%

Artificial Analysis benchmarks aa-v5

Current Intelligence Index components

AA-Briefcase v1.1
1485.1 Elo
AutomationBench-AA
77.5%
private 657-task split from dataset v1.0.6
GDP.pdf
21.8%
FrontierCode v2.0
25.7%
AA Stirrup · 132 tasks, 3 attempts per task
Terminal-Bench v4.0
57.1%
AA mini-swe-agent standardized runs · 66 tasks, pass@1 averaged over 3 repeats per task
AA-Omniscience v1.1
46.1 native net
HLE-Diamond
56.7%
CritPt v2 Hard
52.5%
AA Stirrup · hard subset, 40 of 70 problems
Terminal-Bench-Science v0.1.0
52.9%

Standalone evaluations

GDPval-AA v2.1
1633.5 Elo

Historical evaluations

SciCode v1.0.1
61.8%
AA Long Context Reasoning v1.1
79.7%
Humanity's Last Exam
57.1%
CritPt
27.1%
FrontierCode
43.5%

Similar models

Support me! Patreon