Google unveiled the Gemini 4 Neon model, claiming superiority over key competitors — Anthropic's Claude Opus 5.5 and OpenAI's GPT-6.2 Sol — in most benchmarks. However, in a number of tasks the rivals proved stronger, and this changes the balance in the race among AI leaders.

The announcement of the new model began on November 18. Gemini 4 Neon is positioned as a cutting-edge solution: according to the developers, it quickly completes complex tasks in programming, office work, and scientific calculations. The launch was accompanied by a comment from Demis Hassabis — Nobel laureate, co-founder and chairman of Google DeepMind, and chief scientific officer of Alphabet. He highlighted the model's speed and noted that Pro and Ultra subscribers can already use it.

Where to get access

The model is being rolled out gradually, so some users will not see it right away. Access is opening through several channels:

  • for developers — via the Gemini API in Google AI Studio;
  • for companies — in the Gemini for business service;
  • for paid Google AI Pro and Ultra subscribers — in Google Antigravity, a development environment with AI agents;
  • for Google AI Pro and Ultra subscribers — in the Gemini app.

The balance of power across 18 tests

Google compared the three models across 18 benchmarks. It is important to keep in mind: this is the company's own data, and independent checks will show how accurate it is. Gemini 4 Neon took first place in 15 tests, Claude Opus 5.5 won in two, and GPT-6.2 Sol in one.

Gemini 4 Neon showed the largest gap in working with huge volumes of data. In the GraphWalks test, the model must navigate text up to a million tokens long — hundreds of pages at once. Gemini 4 Neon's result is 95.2%, GPT-6.2 Sol's is 80.1%, and Claude Opus 5.5's is 66.8%.

The newcomer also leads noticeably in office work automation: in AutomationBench it scores 56.3% versus 42.5% for Claude and 41.3% for GPT. It is also ahead in scientific and mathematical tasks, in complex reasoning, and in vulnerability discovery. In the WebExploitBench test, where AI looks for weak spots in websites, Gemini 4 Neon scored 51.3% versus 37.1% for Claude.

The competitors' strengths

The rivals also have their own trump cards:

  • Claude Opus 5.5 slightly outpaced Gemini in Terminal-bench 4.1 and FrontierCode 2.0 — tests on working in the command line and solving complex programming tasks;
  • GPT-6.2 Sol is the best at understanding graphs and charts — in the Chartography test it scores 75.1% versus 73.5% for Gemini;
  • GPT-6.2 Sol took second place in long-video analysis and working with large texts.

The gap in the tests where Claude wins is about one percentage point. Therefore, for programming the models can be considered almost equal. For working with long documents and automating routine tasks, judging by Google's figures, Gemini 4 Neon is better suited.

Cryptalist analysis: For investors tracking the AI sector, this is a signal of intensifying competition between Alphabet, OpenAI, and Anthropic: the leader's advantage in many tests is measured in just a few percent, which means the market is entering a phase of rapid mutual catch-up. No sustained dominance by a single platform is forming — the bet should be on diversifying ecosystems rather than on a single vendor.