Neon Field GuideIndependent reference

SOURCE REVIEW · NOV 20, 2026

What do Gemini 4 Neon benchmarks measure?

DeepMind’s November model card reports evaluations for coding, reasoning, knowledge work and multimodal understanding. Match the benchmark to your task before using a result to choose a model.

Selected results from the November 18 model card

DeepMind-reported results. This site has not reproduced these runs. A dash indicates that the source table does not report a score.

BenchmarkNeonClaude Opus 5.5GPT-6.2 SolSource
AA Index v5.0.1615957DeepMind · Nov 18
Vals Index v2.270.9%70.1%66.1%DeepMind · Nov 18
CursorBench 4.165.4%58.1%—DeepMind · Nov 18
Terminal-bench 4.1 (adjusted)65.2%65.8%61.5%DeepMind · Nov 18
SWE-sweep (provider harness)39.0%10.1%8.3%DeepMind · Nov 18
SWE-sweep (Mini-SWE harness)12.8%3.6%11.6%DeepMind · Nov 18
AutomationBench56.3%42.5%41.3%DeepMind · Nov 18
RiemannBench84.0%80.0%80.0%DeepMind · Nov 18
Chartography73.5%66.3%75.1%DeepMind · Nov 18
LVBench89.0%83.7%87.6%DeepMind · Nov 18
GraphWalks (256k–1M, BFS F1)95.2%66.8%80.1%DeepMind · Nov 18

GPT-6.2 Sol leads Neon on Chartography in this table. The two SWE-sweep rows also show why the harness should be recorded alongside the score. Use a task-specific evaluation before choosing a model for production.

Choose evidence for your workload

Repository changes
Start with CursorBench and SWE-sweep. Check the agent harness because tooling can affect the outcome.
Terminal workflows
Review Terminal-bench and check which metric is reported. Record your own commands, timeouts and manual interventions.
Charts and documents
Look at Chartography for chart understanding. For production extraction, test the exact documents and schema you use.

Run a fair comparison

Keep prompts, tools, test fixtures and acceptance criteria fixed. Record the model ID, thinking level and date. Separate wall-clock time from review time, and compare cost per accepted result. Keep failed and retried runs in the log.

Download the benchmark worksheet (CSV)

Source record

Reviewed November 19, 2026. The model card is the source for benchmark coverage. This site has not run or reproduced the published scores.

Primary sources

Google launch announcement · DeepMind Neon model card