Introducing Gemini 4 Neon
Google’s Gemini 4 Neon tops the WebExploitBench v1.2 penetration-testing benchmark at half the cost of competing large models.
12 comments
After a bit of testing it feels a lot more benchmaxxed than Carbon unfortunately, it's much slower and more expensive (like 3-5x slower and more expensive) than GPT-6.2 Sol and roughly the same quality on a couple of our internal evals like summarization and visual slide deck QA.
Neon is an impressive model. It ranks 3rd on the AAI index, which is quite a hard benchmark. The fact that it performs better than Carbon shows a continuing trend among top labs where the adapted model usually outperforms the base model across all major benchmarks.
I (biasedly) enjoy all the Gemini models, especially because their outputs feel distinctly different from Claude/GPT models. My honest thoughts though are that Hydrogen is so fast and smart that it still is better as a daily driver next to Opus 5.5, and Carbon/Fable are best used for tough/longer-running tasks that require a lot of thinking.
I’m personally a big fan of Carbon’s personality and writing style I find myself defaulting to it quite often!


Somehow DeepMind went from shipping nothing but flash models for almost a year to big model releases every week.
This is a smaller model than Carbon but beats Carbon on AA Index (it has an xhigh thinking level where Carbon only goes up to high), priced $3/M input tokens and $15/M output tokens, doesn't have the 50% off discount so Carbon is actually cheaper through the end of November.