Atlas

Models

Google

Gemini 4 Neon

170.1
Capability · 95% interval 168.1–172.5
Rank 3 of 901 (interval 1–4)
Relative cost 3.1× · $0.34 per short-answer task · $1.95 per agentic task

Gemini 4 Neon (`gemini-4-neon`) is Google's Gemini 4 model for complex coding, analytical reasoning and production enterprise workflows, adapted from Gemini 4 Carbon and positioned between the frontier Carbon and the fast Hydrogen. Google released it on November 18, 2026 as a stable model through the paid Gemini API, Google AI Studio, Gemini for business, Google Antigravity and the Gemini app, at $3 per million input, $0.15 per million cached input and $15 per million output tokens, with a cyber version rolling out to Fairwind Program defenders in the following days. It accepts text, image, audio, video and PDF input and returns text, with a unified 1,048,576-token input-plus-output context window, 65,536 default synchronous output tokens or up to 1,048,576 with long decode, a March 2026 knowledge cutoff, and low, medium, high (the default) and xhigh thinking levels.

Scored on 301 benchmarks · 379 total scores · 369 independently reported

Score providers

Organizations reporting headline scores for 301 benchmarks

Gemini 4 Neon benchmark scores

In fitin the capability model (58 of 301)
Professional · 135 benchmarks · first place on 30
Vals Finance Indexx-high
1 / 42———
Vals Legal Indexx-high
1 / 37———
Legal Research BenchIn fitx-high
1 / 88~2 mo~Feb 2028—
Legal Research Bench Administrative/Regulatoryx-high
1 / 90———
Legal Research Bench Constitutional/Civil Rightsx-high
1 / 90———
Software · 57 benchmarks · first place on 10
CursorBench 4.1In fitx-high
1 / 12~1 mo~Feb 20282026-11-18
Bug Hunt BenchIn fitx-high
1 / 49~2 mo~Dec 20272026-11-19
Terminal-Bench 4.0 — MLx-high
1 / 61———
SEC-Bench Pro - release-reported cohortIn fitx-high
1 / 17~1 moSep 2026*2026-11-18
SWE-sweepIn fitx-high
1 / 16~2 mo~Oct 20282026-11-18
Language & writing · 32 benchmarks · first place on 6
LMArena Text Arena Hard Prompts Elohigh
1546Arena
1 / 256——2026-11-18
LMArena Text Arena Chinese Elohigh
1604Arena
1 / 244——2026-11-18
Arena Text — Instruction Following (No Style Control)high
1542Arena
1 / 256——2026-11-18
Arena Text — Creative Writing (No Style Control)high
1534Arena
1 / 255——2026-11-18
Arena Text — Hard Prompts (No Style Control)high
1553Arena
1 / 256——2026-11-18
Tool & computer use · 21 benchmarks · first place on 12
AutomationBench-AA 1.0.6In fitx-high
1 / 167~2 mo~May 2027—
AutomationBench 1.0.6 (Zapier)In fitx-high
1 / 43~2 mo~Sep 20272026-11-18
AutomationBench 1.0.6 Salesx-high
1 / 2——2026-11-18
AutomationBench 1.0.6 Marketingx-high
1 / 2——2026-11-18
AutomationBench 1.0.6 Operationslow
1 / 2——2026-11-18
Science · 15 benchmarks · first place on 5
LMArena Text Arena Occupational Life, Physical, and Social Science Elohigh
1550Arena
1 / 255——2026-11-18
Arena Text — Life, Physical & Social Science (No Style Control)high
1564Arena
1 / 255——2026-11-18
Terminal-Bench Science — Physical Sciences (Vals)x-high
1 / 46———
Terminal-Bench Science — Earth Sciences (Vals)x-high
1 / 46———
Artificial Analysis Engineering Indexx-high
1 / 26———
Knowledge & reasoning · 10 benchmarks · first place on 1
AA-Omniscience v1.1 Indexx-high
1 / 85———
GraphWalks BFS: 256K to 1M Aggregatex-high
2 / 11——2026-11-18
Humanity's Last Exam Diamond (Knowledge Partition)x-high
2 / 109———
SimpleBenchIn fitnot reported
3 / 114~4 moSep 20262026-11-19
Humanity's Last Exam Diamondx-high
3 / 153———
Health & medicine · 9 benchmarks · first place on 3
LMArena Text Arena Occupational Medicine & Healthcare Elohigh
1538Arena
1 / 245——2026-11-18
Arena Text — Medicine & Healthcare (No Style Control)high
1543Arena
1 / 245——2026-11-18
Artificial Analysis Healthcare & Medical Indexx-high
1 / 26———
DAYJOB: Healthcare (October 2026 rescoring)In fithigh
2 / 45~1 mo~Sep 2027—
AA-Omniscience Health Accuracyx-high
3 / 26———
Math · 8 benchmarks · first place on 4
OTIS Mock AIME 2024-2025In fitx-high
1 / 143~1.1 yearsJul 2025—
LMArena Text Arena Occupational Mathematical Elohigh
1544Arena
1 / 244——2026-11-18
Arena Text — Occupational Mathematical (No Style Control)high
1557Arena
1 / 244——2026-11-18
RiemannBench (November 2026 rescoring)In fithigh
1 / 26~2 mo~Mar 2027—
FrontierMath Tier 4 (v2) - PrivateIn fitx-high
5 / 76~3 moFeb 2026—
Perception & media · 5 benchmarks · first place on 1
Humanity's Sixth SenseIn fitx-high
1 / 37~2 mo~Sep 2027—
ChartographyIn fitx-high
2 / 49~2 mo~Jun 20272026-11-18
LVBenchIn fitx-high
2 / 72~7 mo~Apr 20272026-11-18
Furniture Assembly BenchmarkIn fitx-high
2 / 42~2 moNov 2026—
Chartography (November 2026 rescoring)In fithigh
6 / 36~3 mo~Dec 2026—
Composite indexes · 5 benchmarks · first place on 3
Vals Index v2.2x-high
1 / 37———
Artificial Analysis Economics Indexx-high
1 / 26———
LiveBench 2.0 (2026-11-06 release)In fitx-high
1 / 36~1 mo~Mar 2027—
Artificial Analysis Intelligence Index v5.0.1x-high
2 / 76———
Capabilityx-high
170.1 / 171Capability
3 / 901——2026-11-20
Games · 2 benchmarks · first place on 1
Chess Puzzles (task version 2.0)In fitx-high
1 / 92~4 mo~May 2027—
Mystery Game PuzzlesIn fitx-high
3 / 84~4 moSep 2026—
Business simulations · 1 benchmark · first place on 1
Vending-Bench 2not reported
1 / 70——2026-11-19
Safety & alignment · 1 benchmark · first place on 1
Gray Swan IPI Attack Success (k=15)x-high
1 / 30——2026-11-18

Gemini 4 Neon head to head

Related models

Loading Atlas data…