Model results · Google

Gemini 4 Neon

Suitev4
Gemini 4 Neon across every effort level
In Antigravity across four effort levels (no Max): combined score 84.32 at Low, 87.48 at Medium, 87.87 at High and 87.70 at Extra-high. 17 of 23 tasks passed at Low and all 23 above; Code quality 66.15 to 69.07; 5.1 to 10.1 minutes per task. Low and Extra-high are judged on 22 runs: Grok 4.6’s probe of codeccore returned malformed JSON twice.
November 21, 2026 · 23 tasks · 4 effort levels · 92 runs of Gemini 4 Neon · VulcanBench Frontier v4 · Code quality protocol v3.32
→
Suitev4
Beside Claude Opus 5.5 and GPT-6.2 Sol
From Medium up Gemini 4 Neon sits 1.0 to 1.5 points behind GPT-6.2 Sol and about 3 behind Claude Opus 5.5, at $0.92 to $1.80 a task against $0.33 to $0.46 and $2.84 to $4.11. On the Frontier v4 board its High level ranks 43rd of 95 columns.
November 21, 2026 · comparison card · VulcanBench Frontier v4
→
Suitev4
Cost and tokens across effort levels
$0.73 to $1.80 per task at list API rates and 0.91M to 2.45M raw tokens per task across the four effort levels; the whole sweep prices at $106.84 for 92 runs, $1.16 per task.
November 21, 2026 · API-equivalent estimates at list rates, solver inference only · VulcanBench Frontier v4
→
Report02
How Gemini 4 Neon thinks
A trace analysis of all 92 runs: when Neon thinks, reads, experiments and edits, how that shifts with effort, why Low misses six tasks, and how it compares with Claude Fable 5.2.
Reports · November 21, 2026
→
Tracesv4
Redacted agent traces: Gemini 4 Neon on Frontier v4
The trace repository for the sweep in the report above, published separately on GitHub. Structure only: every model response with its token counts, every tool call with its timing, and a closing summary per run, without the task content.
GitHub · vulcanbench-gemini4-neon-traces
→

How would Gemini 4 Neon do on your stack?

Public results answer "which model is stronger on this suite." Whether Gemini 4 Neon belongs in your routing table depends on your languages, your codebases, and your task mix, that gets measured, not guessed. Grading rules are in the methodology.