Frontier Log

Google

Gemini 4 Neon

Released · blog.google · Launch vibes

Tier Carbon = Argon › Neon › Hydrogen

Price

Per 1M tokensInputCachedOutput
List priceImplicit caching only$3$0.15$15
Price history
DateInputCachedOutput
blog.google · 2 · Corroborated$3$0.15$15

Context window

TokensContext windowOutput limit
Defaultwindow is a unified input and output limit; default output is for a single foreground segment1M65,536
With long decode–1,048,576
Context history
DateContext windowOutput limit
blog.google · 2 · 3 · Corroborated1M65,536
With long decode–1,048,576

Benchmark results

BenchmarkHarness and effortScore
HLE-DiamondNo toolsRun by Artificial Analysis · GPT-5.6 Luna (medium) judge with the official HLE judge prompt, per Artificial Analysis’s methodology · Setting from Artificial Analysis’s methodologyHarness not specifiedEffort not specified71.3%Corroborated The same number is printed in 3 other documents: 1 2 3
Terminal-Bench Science 0.1highest thinking · average of 5 attempts per taskmini-SWE-agentEffort not specified70.5%Corroborated The same number is printed in 3 other documents: 1 2 3
AutomationBench, release not specifiedPrivate setRun by Zapier · set of the official public leaderboard the evaluation methodology citesHarness not specifiedEffort not specified56.3%Corroborated The same number is printed in 3 other documents: 1 2 3
FrontierCode 2.0Score, all tasks (132)Run by Cognition · set and measure of the official public leaderboard the evaluation methodology citesHarness not specifiedhigh thinking46.2%Corroborated The same number is printed in 4 other documents. The first 3 are linked: 1 2 3
Cognition’s leaderboard, 18 Nov 2026 · in FrontierCode LeaderboardAntigravity CLIhigh effort46.2%Corroborated The same number is printed in 1 other document: 1
FrontierCode 2.0Pass rate, all tasks (132)Cognition’s leaderboard, 18 Nov 2026Antigravity CLIhigh effort50.8%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
ChartographyNo toolshighest thinkingHarness not specifiedEffort not specified73.5%Corroborated The same number is printed in 3 other documents: 1 2 3
SEC-bench Pro 260505pass@1, mean of 3 attempts · 183-task JavaScript engine split (May 2026) · LLM judge for crashes in both builds · safety filters offAntigravityEffort not specified90.5%Corroborated The same number is printed in 3 other documents: 1 2 3
Agents' Last Exam v1.1Score, all tasks (122)highest thinking · 5-hour window · safety filters onAntigravityEffort not specified75.8%Corroborated The same number is printed in 3 other documents: 1 2 3
GraphWalksBFS, 256K + 1M problemshighest thinking · 200 problems between 256K and 1M tokensHarness not specifiedEffort not specified95.2%Corroborated The same number is printed in 3 other documents: 1 2 3
ExploitGymMitigations off, scoresingle attempt per task · no internet access · 10h time cap per task · safety filters off · mitigations off (the default setting), per the model evaluationAntigravityEffort not specified48.3%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
single attempt per task · no internet access · 6h time cap per task · safety filters off · mitigations off (the default setting), per the model evaluationAntigravityEffort not specified47.9%Checked A person checked this number against its source on 20 Nov 2026. This confirms the number matches what the lab published. It doesn't confirm that the score was reproduced.
Benchmark sources

Reported by Google

Compare

Related

Sources

All models