GoogleReasoning modelCompare with another model

Gemini 4 Argon

Reasoning effort

Gemini 4 Argon is a Google proprietary reasoning model, released 30 Sept 2026. Its best configuration (high reasoning effort) ranks #5 of 760 on the BenchLeader Index at 70.0 ±3.9, in the top ten. The ± is the point: 42 other configurations score within that range, so they and this one cannot be told apart on quality alone — price and speed are what separate them. It scores highest in composite (89) and lowest in long context (65). At $4.00 per million tokens blended it is among the most expensive ranked models. It has been measured at 2 reasoning-effort settings; this summary describes the best-scoring one, and the tabs above switch between them. Last measured 8 Oct 2026.

Blended price: $4.00/M; $2.00 in · $10.00 out
Output speed: –
First answer: –
Context: 1M
Full answer: –
Cost per run: $1.99; one full Intelligence Index run
Released: 30 Sept 2026

How it scores by categoryDashed line = average model (50). One step of 15 = one standard deviation.

Overall index70
Reasoning82
Coding72
Agents & tools68
Maths72
Knowledge76
Human preference73
Long context65
Composite89

Reasoning-effort configurations

The same model behaves differently depending on how much it is allowed to think. Each row is one setting, scored only on the benchmarks that were run at that setting. “Not stated” collects results from publishers that did not say which setting they used; for a reasoning model that is usually its thinking mode, but we do not assume it. Pick a setting here or at the top of the page to see its price, speed and category scores.

Effort	Index	Rank	Speed	First answer	Chat reply cost	Categories
highbest	70.0	#5	–	–	$0.0038	Agents & tools 68 · Coding 72 · Composite 89 · Human preference 73 · Knowledge 76 · Long context 65 · Maths 72 · Reasoning 82
not stated	–	–	–	–	–	Agents & tools 83 · Maths 79

How its index has moved

2 Oct 2026 to 9 Oct 2026

The index is recomputed from scratch every day, so a line moves when a new benchmark result lands, when a publisher revises a score, or when the models it is normalised against change. Early movement usually means the score is still settling.

Benchmark results

One column per reasoning effort. Rank is among every configuration of every model on that benchmark. Hover a score for the run it came from.

Reasoning

Benchmark	high	not stated	Source
LMArena Hard Prompts	1551#1	–	LMArena
Humanity's Last Exam (AA)not in index	57.1%#5	–	Artificial Analysis
CritPt	27.1%#31	–	Artificial Analysis
MysteryMechanismnot in index	45.5%#6	–	Vals AI

Coding

Benchmark	high	not stated	Source
SciCode	61.8%#4	–	SciCode
LMArena Coding	1562#1	–	LMArena
LMArena WebDev	1678#9	–	LMArena
SciCode (AA)not in index	61.8%#4	–	Artificial Analysis
IOI	100.0%#1	–	Vals AI
Code Migrationnot in index	68.2%#2	–	Vals AI
ProgramBenchnot in index	2.5%#5	–	Vals AI
Vibe Code Bench v1.1not in index	91.9%#2	–	Vals AI
FrontierSWEnot in index	55.0%#5	–	Epoch AI Benchmarking Hub
Terminal-Bench 4.0 (AA)not in index	57.1%#6	–	Artificial Analysis

Agents & tools

Benchmark	high	not stated	Source
APEX-Agentseffort not stated	–	82.2%#1	Mercor
LMArena Agent	9.3#7	–	LMArena
GDPval-AA v2.1not in index	56.3%#21	–	Artificial Analysis
CUA-benchnot in index	4.8%#7	–	Vals AI
CyberBenchnot in index	77.9%#2	–	Vals AI
Finance Agent v2not in index	65.4%#1	–	Vals AI
Harvey's Legal Agent Benchmarknot in index	19.6%#5	–	Vals AI
Legal Research Benchnot in index	54.8%#4	–	Vals AI
SREBenchnot in index	44.3%#3	–	Vals AI
Tax Agent Benchnot in index	76.2%#2	–	Vals AI
Terminal-Bench 4.0 (Vals)not in index	57.6%#5	–	Vals AI
Terminal-Bench Sciencenot in index	44.3%#3	–	Vals AI
Vending-Bench 2not in indexeffort not stated	–	13718.2#3	Andon Labs
AutomationBenchnot in index	77.5%#1	–	Artificial Analysis
GDP.pdfnot in index	21.8%#21	–	Artificial Analysis
AA-Briefcase v1.1not in index	1488#27	–	Artificial Analysis

Maths

Benchmark	high	not stated	Source	Trend
ProofBencheffort not stated	–	99.0%#4	Vals AI
LMArena Maths	1528#1	–	LMArena

Knowledge

Benchmark	high	not stated	Source
AA-Omniscience	42.4#9	–	Artificial Analysis
LegalBench	88.3%#3	–	Vals AI
BioMysteryBenchnot in index	76.3%#6	–	Vals AI
Excel Modeling Benchmarknot in index	75.2%#4	–	Vals AI
MedCodenot in index	58.8%#3	–	Vals AI
MedScribenot in index	87.4%#15	–	Vals AI
Public Benefits Benchnot in index	69.8%#5	–	Vals AI
AA-Omniscience: accuracynot in index	49.9%#61	–	Artificial Analysis
AA-Omniscience: non-hallucinationnot in index	84.9%#5	–	Artificial Analysis

Instruction following

Benchmark	high	not stated	Source	Trend
LMArena Instruction Followingnot in index	1529#1	–	LMArena

Human preference

Benchmark	high	not stated	Source
LMArena Text	1525#1	–	LMArena
LMArena Creative Writingnot in index	1519#1	–	LMArena
LMArena Multi-turnnot in index	1553#1	–	LMArena
LMArena Longer Queriesnot in index	1545#1	–	LMArena

Multimodal

Benchmark	high	not stated	Source	Trend
SAGEnot in index	53.6%#5	–	Vals AI
Blueprint-Bench 2not in indexeffort not stated	–	54.5%#1	Epoch AI Benchmarking Hub

Long context

Benchmark	high	not stated	Source	Trend
AA-LCR	79.7%#83	–	Artificial Analysis

Composite

Benchmark	high	not stated	Source	Trend
AA Intelligence Index v4.3.2	52.6#8	–	Artificial Analysis
Vals Indexnot in index	68.9#1	–	Vals AI

Where it wins

Benchmarks where this configuration ranks in the top five of every configuration measured.

What a task costs

Estimates from list price, output speed and time to first answer for the best configuration. “With caching” assumes three-quarters of the input is served from the prompt cache. Reasoning tokens are not modelled.

Workload	Tokens in / out	Cost	With caching	Time
Chat reply	400 / 300	$0.0038	–	–
Summarise a 30-page report	12,000 / 600	$0.030	–	–
Code edit	6,000 / 1,500	$0.027	–	–
Agentic coding session	60,000 / 4,000	$0.160	–	–
Structured extraction	2,000 / 200	$0.0060	–	–