Blog

Google Just Shipped a Wave of New Gemini Models. Which One Should You Pick?

Share

Google recently released a series of new models named after chemical elements: Argon, Carbon, Hydrogen and Neon, with Krypton reportedly on the way. They map to different versions and fine-tuned variants of the Gemini 3.1 and Gemma 4 families.

Here's how the four models that are already live compare in overall character, coding ability and writing ability.

1. Carbon: the all-round flagship and deep-reasoning champion

(Pro / Ultra tier, or a Thinking model)

In chemistry, carbon is the basis of life. In this lineup, Carbon is the flagship with the strongest overall ability, the most parameters or the deepest reasoning chain.

  • Core traits: Exceptionally strong logical reasoning and very high knowledge density. On complex tasks it shows clear chain-of-thought behavior and excellent planning, but its time to first token (TTFT) is relatively slow.
  • 💻 Coding (⭐⭐⭐⭐⭐): The strongest coder of the four, especially for complex engineering. It doesn't just write code; it proactively considers edge cases and performance.
    • Best for: building complex system architectures from scratch, debugging across many files, tough competitive programming problems (LeetCode Hard level).
  • ✍️ Writing (⭐⭐⭐⭐)
    • Best for: in-depth industry reports, polishing academic papers, rigorous long-form writing. Its style is precise and highly structured, but for literary writing that needs flair and feeling it can come across as stiff, with a heavy "AI literature review" feel.

2. Argon: specialized for creative writing and human preference

(A high-EQ fine-tune of the main Pro tier)

🤚 Gemini 4 Argon is arguably the current best answer for AI fiction writing!

Argon is an inert gas, but in testing, Argon shows extremely high alignment and an excellent ear for language. It's very likely a full-strength model deeply optimized for human preference (RLHF) and humanities tasks.

  • Core traits: High EQ and very strong instruction following. It's great at reading the user's unstated intent. Output is well formatted and comfortable to read.
  • ✍️ Writing (⭐⭐⭐⭐⭐): The best writer of the four.
    • Best for: continuing fiction, marketing copy, viral social media posts and newsletters, role-play (RP), complex emails. Its prose is very natural, rarely stilted or full of mechanical "first, second, finally" structures, and it imitates specific styles and tones well.
  • 💻 Coding (⭐⭐⭐⭐)
    • Best for: everyday full-stack development, front-end UI code (React/Tailwind and the like, with good taste), writing comments and docs. It's a step behind Carbon on hardcore algorithms, but excellent for interactive coding where you discuss requirements as you go.

3. Neon: the fast, cost-effective STEM workhorse

(The main Flash-tier model)

Neon is used in neon lights and stands for brightness and speed. Neon corresponds to Google's much-loved Flash series (such as an evolution of Gemini 3.8 Flash), balancing speed, low cost and high intelligence.

  • Core traits: Extremely fast, little filler, straight to the point. While staying lightweight with very high throughput, it inherits most of the flagship's STEM strength.
  • 💻 Coding (⭐⭐⭐⭐☆): The value and efficiency king for everyday coding.
    • Best for: the completion model behind IDEs like Cursor and Windsurf, quick Python scripts, writing regular expressions, converting SQL queries, routine bug fixes. Its fast responses make it ideal for high-frequency pair programming.
  • ✍️ Writing (⭐⭐⭐☆)
    • Best for: meeting notes, news summaries, everyday business documents. Its style is crisp, concise and practical, not suited to ornate or emotionally nuanced writing.

4. Hydrogen: ultra-light and built for the edge

(Flash-Lite / open Gemma tier)

Hydrogen is element number one and the lightest. Hydrogen is clearly a small, lightweight model (around the 8B/9B class, or a Flash-Lite version).

  • Core traits: Tiny, with millisecond-level responses. Its knowledge isn't as deep as the other three, and it can hallucinate on very long, complex logic, but it handles single, well-defined small tasks with ease.
  • 💻 Coding (⭐⭐⭐)
    • Best for: simple syntax conversion (e.g., turning JSON into a TypeScript interface), basic command lookups (how to write a Git or Linux command), simple single-function generation. Not recommended for complex business logic.
  • ✍️ Writing (⭐⭐⭐)
    • Best for: text classification, sentiment analysis, simple translation, rewriting with fixed templates (e.g., making a sentence more polite).

💡 Summary: how to choose for writing and coding

Use these rules of thumb to match the model to the job.

For writing tasks

  1. Creative writing, copywriting, anything that needs a human feel and good prose 👉 Argon (best ear for language, least "AI flavor").
  2. Long in-depth research reports, rigorous technical docs, literature reviews 👉 Carbon (tight logic, good with huge amounts of context).
  3. Quick summaries, routine formulaic emails 👉 Neon or Hydrogen (instant results, efficiency first).

For coding tasks

  1. Hard problems, architecture design, weird bugs 👉 Carbon (highest ceiling, deepest reasoning).
  2. High-frequency everyday business code, front-end work, fast in-editor completion 👉 Neon (blazing fast, very accurate, best overall feel).
Google Just Shipped a Wave of New Gemini Models. Which One Should You Pick? - SoloEnt